Curriculum alignment · The course
Grade 10: units, topics and lessons
Every Grade 10 topic in the LittleAIMaster AI course, its lessons, and the requirements from 14 AI curricula it helps meet. Open a topic to see its lessons; lessons cited as evidence are in bold.
6units75topics1,934lessons267requirements met
Looking for the learner view? See the Grade 10 learning path.
Unit 1: Neural Network Basics 12 topics
Welcome to Grade 10!24 lessons· meets 10 requirements inUAECBSECISCE + statesEngland + Cambridge+1
Lessons
- Welcome to Grade 10!
- What Are Neural Networks?
- Predict the Node
- Why Did the Lab Break Through?
- The Three Layers
- What is Deep Learning?
- From Brain Idea to AI Revolution
- Build a Network!
- Deep Learning or Not?
- Architecture Types
- ML vs Deep Learning
- Deep Learning Potential
- Object Detection in Google Photos
- Sort the Differences
- How Networks Learn
- One Training Step
- Activation Functions
- Line or Curve?
- Neural Network Challenge
- Chapter Quiz
- Explain Neural Networks
- Deep Learning Vocabulary
- Byte's Roadmap
- AI Scientist Unlocked!
Requirements this topic helps meet
- UAENeural network design and performance
- CBSEDifferentiate AI, ML, and DL and common data terminologies; familiarise with supervised, unsupervised, and reinforcement learning-based approaches; understand subcategories (classification, regression, clustering, association, ANN, CNN); understand neural networks and how AI makes a decision (Unit 2: Advanced Concepts of Modeling in AI).
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain the brief history and primary elements of AI including Machine Learning (ML) vs Deep Learning (DL)/neural networks, and applications of ML/DL such as image recognition, computer vision, speech recognition and information retrieval
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
86 Billion Neurons!22 lessons· meets 10 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- 86 Billion Neurons!
- How Real Neurons Work
- Why Did Scientists Copy Neurons?
- How Brains Learn
- Wire Together?
- A Brain Rule Became AI History
- The Artificial Neuron
- Run One Neuron
- Biology vs Artificial
- Brains Are Efficient
- Brain vs Computer
- Neuron Signal Flow
- Neuron Learning Strength
- Face Grouping in Apple Photos
- Spam Detection Neuron
- Score the Email
- Spot the Neuron Features
- Neuron Inspiration Quiz
- Explain the Inspiration
- Neuron Vocabulary
- The Power of Networks
- Neuron Explorer!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from giving examples of intelligent vs. non-intelligent machines and discussing what makes a machine 'intelligent' (K-2), to demonstrating a machine perception tool (e.g. speech transcription, object recognition) and explaining why it is perception rather than mere sensing (3-5), to giving examples of specialized perception algorithms (face detection, obstacle detection, speech recognition, etc.) (6-8), to explaining how perception algorithms power real-world applications like license plate readers and Google Lens (9-12).
- AI4K12 (US)Progress from constructing a map of a home, school, or neighborhood and understanding that 'the map is not the territory' (K-2), to demonstrating how different map styles (geometric vs. topological) capture different aspects of the world (3-5), to showing how a game board (tic-tac-toe, chess) can be represented as a plain-language description of squares, relationships, and piece positions (6-8), to describing how to represent a concept as a schema.org-style schema with superconcepts and inherited properties (9-12).
- AI4K12 (US)Progress from describing and giving examples of how people vs. computers learn (K-2), to differentiating human flexible learning from computer pattern-finding/trial-and-error learning (3-5), to contrasting human learning characteristics with narrowly-defined machine learning problems (6-8), to defining supervised, unsupervised, and reinforcement learning and giving human-learning analogues for each (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
The First Learning Machine!26 lessons· meets 11 requirements inCBSECISCE + statesEngland + CambridgeCSTA (US)+1
Lessons
- The First Learning Machine!
- The Perceptron Formula
- Run the Perceptron
- Why Could One Neuron Not Do Everything?
- Drawing Decision Lines
- The Learning Rule
- A Single Line Can Still Be Powerful
- Build an AND Gate
- One Update Step
- Learning Rule Match
- The XOR Problem
- Line or No Line?
- Perceptron Success Meter
- Spam Filtering in Gmail
- Perceptron Can or Can't?
- Training Algorithm Order
- Logistic Regression: Yes/No with a Probability
- Perceptron or Logistic?
- Run the Sigmoid
- The AI Winter
- Hype or History?
- Spot Perceptron Uses
- Perceptron Quiz
- Explain XOR
- Perceptron Vocabulary
- Perceptron Pioneer!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesDefine artificial intelligence, give a brief history since John McCarthy coined the term in 1956, and explain the Turing Test and its importance
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Debate what differences do or should exist between human and artificial intelligence, sentience, consciousness, rights, and responsibilities.
- CSTA (US)Students study the history of computing and AI, emerging technologies, human-AI relationships, and responsible decision-making about technology use.
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating ways intelligent agents don't understand things the way people do, e.g. no frustration at repeated questions (K-2), to demonstrating what a 'broad AI'/general-AI system could do that current narrow AI cannot (3-5), to defining criteria for consciousness (self-awareness, personal history) and evaluating AI systems or fictional AI characters against them (6-8), to debating alternative perspectives on human vs. artificial intelligence, including the Turing test and artificial superintelligence (9-12).
The Secret Ingredients!23 lessons· meets 8 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- The Secret Ingredients!
- Understanding Weights
- Why Did Two Neurons Learn the Same Thing?
- Movie Recommender Example
- Score a Comedy
- What Bias Does
- Does It Fire?
- Spam Detector Bias
- A Zero Weight Erases a Feature
- Weight & Bias Effects
- Counting Parameters
- Count the Parameters
- Neuron Tuning Meter
- Netflix Preference Signals
- Initialization Matters!
- Weights Learning
- Scale of Modern AI
- Memory Maths
- Spot Parameter Effects
- Weights & Biases Quiz
- Explain Weights and Bias
- Parameter Vocabulary
- Parameters Decoded
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from analyzing a decision-tree-building game and how the learning algorithm adds/replaces nodes (3-5), to comparing how decision-tree learning (add a node) differs from neural-network learning (adjust every weight) (6-8), to describing how decision-tree, neural-net, regression, and reinforcement-learning algorithms each adjust their own type of internal representation (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
The Missing Magic!25 lessons· meets 7 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- The Missing Magic!
- The Linear Problem
- Collapse the Layers
- Line or Bend?
- The Sigmoid Function
- The Tanh Function
- ReLU: The Champion!
- Run the Functions
- ReLU Variants
- Softmax: Multi-Class
- Softmax by Hand
- Activation Matching
- ReLU Helped Networks Scale
- The Decision Guide
- Activation Fit Meter
- Read the Meter
- Translation Output in Google Translate
- Layer Type Sorting
- Comparison Table
- Build a Classifier
- Spot the Activation
- Activation Quiz
- Explain Activation Choice
- Activation Vocabulary
- Activation Expert!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
From LEGO to Networks!27 lessons· meets 7 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- From LEGO to Networks!
- Network Architecture
- Explore Each Layer
- Layer Check
- Dense Layers
- Code Check: One Dense Layer
- Count the Parameters
- Count It Yourself
- XOR, Solved by a Hidden Layer
- Trace the XOR Network
- Architecture Patterns
- Design Decisions
- Architecture Matching
- Architecture Fact Burst
- Task Sorting
- Build a Network
- Parameter Calculator
- Google Photos Needs the Right Shape
- Byte's Architecture Tip
- Design the Network
- Reading Model Summary
- Read the Summary
- Four Common Design Mistakes
- Building Networks Quiz
- Explain the Blueprint
- Network Vocabulary
- Network Architect!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Data's Amazing Journey!25 lessons· meets 7 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- Data's Amazing Journey!
- Forward Propagation
- The Four Steps
- Where Does the Signal Go?
- Single Layer Forward
- Your Turn: One Neuron
- Full Network Forward
- Shape Check
- Trace Example
- Trace It Yourself
- Batch Processing
- Read the Code
- Forward Prop Matching
- Batching Is a Superpower
- Order the Steps
- Common Issues
- Adjust the Signal Flow
- Gmail Scores a Message Fast
- Issue Sorting
- Forward Prop Activity
- Forward Prop Summary
- Forward Prop Quiz
- Explain the Forward Pass
- Forward Prop Vocab
- Propagation Pro!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
AI's Report Card!25 lessons· meets 7 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- AI's Report Card!
- What is Loss?
- Why Loss Matters
- Which Wrong Answer Hurts More?
- MSE: Mean Squared Error
- MAE: Mean Absolute Error
- Compute the Loss
- Loss Encodes Priorities
- MSE vs MAE
- Binary Cross-Entropy
- Rank the Penalty
- Categorical Cross-Entropy
- Choosing the Right Loss
- Confident and Wrong
- Loss Function Matching
- Dial Up the Penalty
- Netflix Learns From Bad Guesses
- Loss Type Sorting
- Loss Properties
- Loss in Code
- Choose the Loss!
- Loss Functions Quiz
- Explain the Loss Function
- Loss Vocabulary
- Loss Master!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Blindfolded on a Mountain!27 lessons· meets 7 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- Blindfolded on a Mountain!
- The Optimization Problem
- The Four Steps
- Which Way Is Downhill?
- The Update Formula
- Do One Update
- Learning Rate
- Small Steps Train Giant Models
- Learning Rate Effects
- Predict the Code
- GD Variants
- Compare the Variants
- Count the Updates
- Advanced Optimizers
- Momentum Check
- GD Matching
- Tune the Descent
- YouTube's Recommender Learns Downhill
- Variant Sorting
- Order the Algorithm
- Byte's Optimization Tip
- Diagnose the Problem!
- Three Traps to Avoid
- Gradient Descent Quiz
- Explain Gradient Descent
- GD Vocabulary
- Optimization Master!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Backpropagation Basics27 lessons· meets 7 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- The Blame Game
- The Gradient Problem
- The Chain Rule
- Who Caused the Error?
- The Forward Pass
- The Backward Pass
- What Does Each Step Multiply By?
- Backprop Steps
- Take the Step
- Speed of Backprop
- Activation Derivatives
- Saved Activations Matter
- Gradient Checking
- Check the Gradient
- Backprop Match
- Trace the Gradient Signal
- Token Errors Flow Backward
- Forward vs Backward
- Algorithm Order
- Byte's Tip: Vanishing Gradients
- Vanishing Signal
- Gradient Detective
- Quick Review: Backpropagation
- Backprop Quiz
- Explain Backpropagation
- Backprop Terms
- Backprop Master!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Your First Neural Network27 lessons· meets 7 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- Build Something Real
- The 10 Steps
- Data Preparation
- Leak or Legit?
- Building the Model
- Count the Weights
- Compiling the Model
- Same Steps, Different Library
- Training Time!
- Count the Updates
- Visualize Training
- Model Evaluation
- Read the Matrix
- Balance the Project Plan
- Teachable Machine Uses the Same Workflow
- Debugging Tips
- Diagnose the Run
- Project Match
- Before vs After
- Project Order
- Byte's Project Tip
- Best Practices
- The Whole Pipeline in Code
- Project Quiz
- Explain the Project Pipeline
- Project Terms
- Pipeline Traced
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Unit Review27 lessons· meets 7 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- What a Journey!
- The Journey Map
- Neuron Recap
- Trace One Neuron
- Why Can High Accuracy Still Be Misleading?
- Activation Review
- The Training Loop
- Which Step Is It?
- Key Formulas
- Run the Formulas
- Myths Busted!
- Diagnose the Model
- Quick Reference
- Pick the Setup
- Tune the Review System
- Face ID Depends on the Full Stack
- Unit 2 Preview
- Concept Match
- Category Sort
- Training Order
- Byte's Unit 1 Tip
- Unit 1 Challenge
- Quick Review: Unit 1 Highlights
- Unit 1 Final Quiz
- Explain How Neural Networks Learn
- Unit 1 Vocabulary
- Unit 1 Champion!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Unit 2: Training Deep Networks 12 topics
Deep vs Shallow Networks25 lessons· meets 10 requirements inADEKCISCE + statesEngland + CambridgeCSTA (US)+2
Lessons
- Welcome to Unit 2!
- Skyscraper Analogy
- Count the Hidden Layers
- Hierarchical Learning
- How Can a Deeper Network Use Fewer Parameters?
- Counting Parameters
- Count the Parameters
- When to Use Each
- Pick the Right Tool
- The Trade-offs
- ResNet Showed Depth Could Go Extreme
- The Revolution
- Depth Through Time
- Myths Busted!
- Choose the Right Depth
- Google Lens Leans on Deep Features
- A Depth Decision Checklist
- Depth Match
- Task Sorter
- Layer Order
- Depth Check
- Depth Quiz
- Explain When Depth Helps
- Depth Terms
- Depth Master!
Requirements this topic helps meet
- ADEKFootprint / resource cost of AI
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Design ways to minimize negative environmental impacts of AI and communicate those ways to others.
- CSTA (US)Students examine security implications and environmental impacts of AI systems within Systems & Security.
- AI4K12 (US)Progress from describing the sounds that make up spoken language and giving example words for each vowel sound (K-2), to illustrating how sequences of sounds are recognized as candidate words despite unclear signals (3-5), to illustrating how sequences of words are recognized as phrases by resolving ambiguity with higher-level information (6-8), to illustrating the full abstraction hierarchy from waveform to sentence in speech understanding (9-12).
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how design and deployment decisions affect an AI system's resource consumption (Engage 5, advanced).
- OECD–EC (Europe)Consider the ethical consequences of AI systems throughout their life-cycle, including environmental impact and societal impact such as platformisation of work and low-cost labelling labour (AI example 57, competence 4.4).
Epochs & Batches25 lessons· meets 6 requirements inCISCE + statesEngland + CambridgeAI4K12 (US)Singapore+1
Lessons
- The Marathon Training Method
- What is an Epoch?
- Is That an Epoch?
- Epoch Exploration
- Why Did One Model Learn So Much Faster?
- What is a Batch?
- Batch Size Trade-offs
- Count the Batches
- Batch Size Secrets
- SGD, Mini-batch or Full?
- Understanding Iterations
- Training Math Challenge
- Predict the Step Count
- Dial Up the Training Load
- ChatGPT Training Relies on Batches
- The Training Loop
- When Should Training Stop?
- Match the Training Terms
- Order the Training Steps
- Batch Size Characteristics
- Training Configuration Check
- Epochs & Batches Quiz
- Explain Epochs, Batches, and Iterations
- Training Vocabulary
- Training Master!
Requirements this topic helps meet
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- SingaporeDevelop device and software operation skills as one of the nine core NDLP digital competencies.
- Korea, Japan, HKUnderstand the mechanisms and characteristics of computers and external devices, internal representation of information, and the limits of computation.
- Korea, Japan, HKUnderstand how information systems store, manage and provide data as services over networks, and evaluate/discuss one's own use of such services.
The Step Size Secret29 lessons· meets 3 requirements inCISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- The Step Size Secret
- The Hill Analogy
- Steps Too Big!
- Why Did the Loss Explode?
- Steps Too Small!
- The Formula
- Run the Update by Hand
- Common Values
- Diagnose the Problem!
- Read the Code
- The Quick Fix
- Lower, Raise or Keep?
- Schedules Intro
- Step Decay
- Step Size on a Real Curve
- GPT-3 Used a Learning Rate Schedule
- Schedule Types
- Schedule Maths
- Adam: Too Small, Typical, Too Big
- Order the Schedule!
- Spot the Problem!
- Checkpoint Quiz
- Explain the Learning Rate Trade-off
- New Vocabulary
- Maya's Next Project
- Key Takeaways
- Finding a Starting Point
- Step Size Master!
- Next Chapter Preview
Requirements this topic helps meet
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
The Hill Strategy29 lessons· meets 3 requirements inCISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- The Hill Strategy
- What's an Optimizer?
- The Evolution
- Updates per Pass
- SGD Explained
- Momentum Magic
- Momentum by Hand
- Momentum Benefits
- Momentum in Python
- RMSprop Idea
- RMSprop by Hand
- Meet Adam!
- Why Adam Wins
- One Adam Step
- Tune the Step Size
- Optimizers at GPT Scale
- The Memory Bill
- Match Optimizers!
- Timeline Order!
- Which Optimizer Is It?
- Common Mistakes
- Spot the Truth!
- Checkpoint Quiz
- Explain Why Adam Is Popular
- New Vocabulary
- Key Takeaways
- Byte's Wisdom
- Optimizer Expert!
- Next Chapter Preview
Requirements this topic helps meet
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
The Ghost Problem29 lessons· meets 5 requirements inCISCE + statesEngland + CambridgeCSTA (US)AI4K12 (US)
Lessons
- The Ghost Problem
- Gradient Refresher
- The Chain Rule
- Why Did the Earliest Layers Freeze?
- Sigmoid Villain
- Visual Example
- One Activation Choice Revived Deep Learning
- ReLU to the Rescue!
- Sigmoid vs ReLU
- Residual Paths Let Gradients Take Shortcuts
- More Solutions!
- Skip Connections
- Skip or No Skip?
- Keep the Gradient Alive
- ChatGPT Uses Gradient-Friendly Design
- Smart Initialization
- Run the Numbers
- Match Solutions!
- History Timeline!
- Sort the Symptoms!
- Spot the Truth!
- Checkpoint Quiz
- Explain Vanishing Gradients
- New Vocabulary
- Diagnose the Training Run
- Key Takeaways
- Byte's Wisdom
- Gradient Master!
- Next Chapter Preview
Requirements this topic helps meet
- CISCE + statesDefine artificial intelligence, give a brief history since John McCarthy coined the term in 1956, and explain the Turing Test and its importance
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Students study the history of computing and AI, emerging technologies, human-AI relationships, and responsible decision-making about technology use.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
The Messy Inputs29 lessons· meets 3 requirements inCISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- The Messy Inputs
- Covariate Shift
- The Big Idea!
- Name the Problem
- The Formula
- Learnable Parameters
- Normalize a Batch
- BatchNorm Benefits
- Train vs Inference
- Which Mode?
- BatchNorm Placement
- Does the Bias Survive?
- Layer Normalization
- Which Direction?
- Stabilize the Layer Inputs
- Google Photos Benefits from Stable Vision Training
- Match Concepts!
- Order the Steps!
- Sort Use Cases!
- Common Mistakes
- Bug or Fine?
- Spot the Truth!
- Checkpoint Quiz
- Explain Why BatchNorm Helps
- New Vocabulary
- Key Takeaways
- What We Still Don't Know
- Normalization Pro!
- Next Chapter Preview
Requirements this topic helps meet
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
The Lazy Student29 lessons· meets 3 requirements inCISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- The Lazy Student
- Overfitting Review
- What is Dropout?
- Diagnose Before You Fix
- Why It Works
- Dropout Rate
- Count the Dropped Neurons
- Train vs Inference
- Inverted Dropout
- Do the Mask Maths
- Where to Apply
- Code Example
- Spot the Misplaced Dropout
- Balance the Dropout Rate
- Pick the Rate
- Spotify Needs Models That Generalize
- Spatial Dropout
- Regular or Spatial?
- Match Concepts!
- Order the Steps!
- Sort by Situation!
- Spot the Truth!
- Checkpoint Quiz
- Explain Why Dropout Helps
- New Vocabulary
- Key Takeaways
- When Dropout Isn't the Answer
- Dropout Expert!
- Next Chapter Preview
Requirements this topic helps meet
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Early Stopping29 lessons· meets 8 requirements inADEKCISCE + statesEngland + CambridgeCSTA (US)+2
Lessons
- The Perfect Moment
- The Overfitting Review
- The Sweet Spot
- Read the Curves
- What Is Early Stopping?
- The Patience Parameter
- Trace the Counter
- Patience Match
- Validation Curves Are the Real Signal
- Restore Best Weights
- Which Weights Do You Keep?
- Early Stopping Steps
- Early Stopping Code
- Predict the Stop
- What to Monitor
- Tune the Stopping Strategy
- Teachable Machine: Choosing Epochs by Hand
- Parameter Sort
- Benefits of Early Stopping
- Potential Drawbacks
- Diagnose the Run
- Regularization Battle
- Combining Techniques
- Spot Early Stopping
- Early Stopping Quiz
- Explain Why Early Stopping Works
- Vocabulary Cards
- Key Takeaways
- Chapter Complete!
Requirements this topic helps meet
- ADEKFootprint / resource cost of AI
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Design ways to minimize negative environmental impacts of AI and communicate those ways to others.
- CSTA (US)Students examine security implications and environmental impacts of AI systems within Systems & Security.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how design and deployment decisions affect an AI system's resource consumption (Engage 5, advanced).
- OECD–EC (Europe)Consider the ethical consequences of AI systems throughout their life-cycle, including environmental impact and societal impact such as platformisation of work and low-cost labelling labour (AI example 57, competence 4.4).
Model Checkpoints29 lessons· meets 3 requirements inCISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- The Crashed Computer
- What Are Checkpoints?
- Why Use Checkpoints?
- Save Point Maths
- Checkpoint Strategies
- The Best-Only Trap
- Strategy Match
- Checkpoint Curiosity
- Keras ModelCheckpoint
- Checkpoint Parameters
- Which Files Get Written
- Parameter Sort
- Loading Checkpoints
- Checkpoint Safety Simulator
- Checkpoints in the Real World
- Resume Training
- Finish the Plan
- File Formats
- Format Match
- Checkpoint Best Practices
- Common Mistakes
- Full Model or Weights?
- Spot the Issue
- Checkpoint Quiz
- Explain Checkpoints
- Vocabulary Cards
- Key Takeaways
- Byte's Message
- Chapter Complete!
Requirements this topic helps meet
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
TensorBoard Basics29 lessons· meets 8 requirements inADEKCISCE + statesEngland + CambridgeCSTA (US)+2
Lessons
- Flying Blind
- What Is TensorBoard?
- TensorBoard Panels
- Which Panel?
- Setting Up TensorBoard
- When Training Becomes Visible
- Setup Order
- Why Is It Empty?
- TensorBoard Callback
- Parameter Sort
- Reading Scalar Graphs
- Diagnose the Run
- Curve Pattern Match
- Custom Logging
- TensorBoard Signal Mixer
- TensorBoard in Practice
- Comparing Runs
- Smooth and Compare
- Weight Histograms
- Read the Histogram
- TensorBoard Best Practices
- Common Issues
- Spot the Issue
- TensorBoard Quiz
- Explain TensorBoard
- Vocabulary Cards
- Key Takeaways
- Byte's Message
- Chapter Complete!
Requirements this topic helps meet
- ADEKFootprint / resource cost of AI
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Design ways to minimize negative environmental impacts of AI and communicate those ways to others.
- CSTA (US)Students examine security implications and environmental impacts of AI systems within Systems & Security.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how design and deployment decisions affect an AI system's resource consumption (Engage 5, advanced).
- OECD–EC (Europe)Consider the ethical consequences of AI systems throughout their life-cycle, including environmental impact and societal impact such as platformisation of work and low-cost labelling labour (AI example 57, competence 4.4).
GPU vs CPU29 lessons· meets 15 requirements inADEKCBSECISCE + statesIB+6
Lessons
- The Waiting Game
- What Is CPU?
- Mystery Speedup
- CPU Superpowers
- Count the Steps
- CPU vs GPU Basics
- When Cores Don't Help
- What Is GPU?
- Parallel Power
- Count the Multiplications
- Sort the Speeds
- Speed Comparison
- The Part That Won't Speed Up
- GPU Memory (VRAM)
- Will It Fit?
- CUDA and Friends
- Processor Choice Simulator
- Hardware Choices in Production
- GPU Terms Match
- Detecting GPU
- GPU Setup Steps
- Cloud GPU Options
- Decision Guide
- Pick the Processor
- GPU vs CPU Quiz
- Explain GPU vs CPU
- GPU Vocabulary
- Byte's Wisdom
- Chapter Complete!
Requirements this topic helps meet
- ADEKFootprint / resource cost of AI
- CBSEAnalyse data (numbers/images), find patterns/relations, and understand the role of Math (statistics, linear algebra, probability, calculus) in AI; understand statistics and probability concepts and apply them to real-life scenarios such as disaster management, sports, disease prediction, weather forecast, traffic estimation (Unit 3: Math for AI).
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- IBExplain how core computer hardware, operating systems, and cloud computing concepts underpin computing systems
- England + CambridgeUnderstand the hardware and software components that make up computer systems, and how they communicate with one another and with other systems
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- England + CambridgeShow understanding of operating system process management (multi-tasking, process states, scheduling algorithms) and virtual memory/paging/segmentation
- CSTA (US)Design ways to minimize negative environmental impacts of AI and communicate those ways to others.
- CSTA (US)Students examine security implications and environmental impacts of AI systems within Systems & Security.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how design and deployment decisions affect an AI system's resource consumption (Engage 5, advanced).
- OECD–EC (Europe)Consider the ethical consequences of AI systems throughout their life-cycle, including environmental impact and societal impact such as platformisation of work and low-cost labelling labour (AI example 57, competence 4.4).
- SingaporeDevelop device and software operation skills as one of the nine core NDLP digital competencies.
- Korea, Japan, HKUnderstand the mechanisms and characteristics of computers and external devices, internal representation of information, and the limits of computation.
- Korea, Japan, HKUnderstand how information systems store, manage and provide data as services over networks, and evaluate/discuss one's own use of such services.
Unit 2 Review28 lessons· meets 3 requirements inCISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- Journey Recap
- Training Fundamentals
- Mystery Review Clue
- Training Concepts
- Count the Updates
- Loss Functions
- Compute the Loss
- Sort the Losses
- Training System Curiosity
- Optimizers Review
- Fitting Problems
- Read the Training Log
- Training Balance Simulator
- Review Concepts in Real Products
- Checkpoints Review
- Model Checkpoints
- TensorBoard Review
- GPU vs CPU Sort
- Training Workflow
- Rapid Review
- Unit 2 Challenge
- Unit 2 Final Quiz
- Explain Unit 2
- Unit 2 Vocabulary
- Debugging a Training Run
- Diagnose the Run
- Byte's Message
- Unit 2 Complete!
Requirements this topic helps meet
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Unit 3: Computer Vision CNNs 12 topics
Introduction to CNNs27 lessons· meets 22 requirements inADEKCBSECISCE + statesIB+5
Lessons
- The Image Problem
- What Is CNN?
- Mystery Image Solver
- What Is Convolution?
- Slide the Filter
- Filters & Features
- CNN Curiosity
- Feature Maps
- Pooling Explained
- Pool It Yourself
- CNN Architecture
- Conv Parameters
- CNN Strength Simulator
- CNNs in Everyday Tech
- Padding Types
- Size and Count
- Famous CNNs
- CNN Applications
- CNN in Keras
- CNN vs Dense
- CNN or Not?
- CNN Challenge
- CNN Quiz
- Explain CNN Basics
- CNN Vocabulary
- Byte's Take
- Chapter Complete!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- ADEKFootprint / resource cost of AI
- CBSEUnderstand and visualize the computer's ability to identify alphabets and handwriting; hands-on Computer Vision unit content (Unit 5: Computer Vision, 20 practical hours).
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Design ways to minimize negative environmental impacts of AI and communicate those ways to others.
- CSTA (US)Students examine security implications and environmental impacts of AI systems within Systems & Security.
- AI4K12 (US)Progress from giving examples of visual features (parts, textures, colors) used to recognize an object class such as cats (K-2), to illustrating how face detection extracts facial features (3-5), to simulating an edge detector to illustrate feature extraction from images (6-8), to explaining how features (formants, edges, intensity gradients) are extracted from waveforms and images at multiple scales (9-12).
- AI4K12 (US)Progress from describing the sounds that make up spoken language and giving example words for each vowel sound (K-2), to illustrating how sequences of sounds are recognized as candidate words despite unclear signals (3-5), to illustrating how sequences of words are recognized as phrases by resolving ambiguity with higher-level information (6-8), to illustrating the full abstraction hierarchy from waveform to sentence in speech understanding (9-12).
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from illustrating why training a classifier for a broad concept like 'dog' requires large, diverse image data by browsing datasets such as ImageNet or Coco (3-5), to illustrating how objects in images are segmented and labeled with bounding boxes to build an object-recognition training set (6-8), to evaluating a real AI training dataset's size, acquisition/labeling method, storage cost, and time-to-produce, including where bias can enter (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how design and deployment decisions affect an AI system's resource consumption (Engage 5, advanced).
- OECD–EC (Europe)Consider the ethical consequences of AI systems throughout their life-cycle, including environmental impact and societal impact such as platformisation of work and low-cost labelling labour (AI example 57, competence 4.4).
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Convolution Operations29 lessons· meets 14 requirements inADEKCBSECISCE + statesIB+3
Lessons
- The Sliding Window
- Meet the Kernel
- Kernel Check
- Convolution Steps
- Try One Window
- Order the Steps
- Convolution Curiosity
- Real Math Example
- Predict the Output
- Famous Kernels
- Kernel Match
- Output Size Formula
- Size It Up
- Output Size Lab
- Convolution in Real Products
- Stride Power
- Padding Explained
- Multiple Filters
- Color Convolution
- Conv2D in Keras
- Count the Parameters
- Parameter Match
- Conv Challenge
- Run Your Sliding Filter
- Convolution Quiz
- Explain Convolution
- Conv Vocabulary
- One Honest Footnote
- Chapter Complete!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- AI4K12 (US)Progress from giving examples of visual features (parts, textures, colors) used to recognize an object class such as cats (K-2), to illustrating how face detection extracts facial features (3-5), to simulating an edge detector to illustrate feature extraction from images (6-8), to explaining how features (formants, edges, intensity gradients) are extracted from waveforms and images at multiple scales (9-12).
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Pooling Layers28 lessons· meets 13 requirements inADEKCBSECISCE + statesIB+3
Lessons
- Too Much Data!
- What Is Pooling?
- Quick Size Check
- Max Pooling, Step by Step
- Why Max Works
- Pool It Yourself
- Average Pooling
- Pooling Sort
- Same Patch, Two Answers
- Global Pooling
- Count the Head
- Pooling Parameters
- Pooling Math
- Size It Up
- Translation Invariance
- Shrink Calculator Lab
- Pooling in Real Vision Systems
- Pooling in Keras
- Predict the Output
- CNN Flow Order
- Strided Convolutions
- Pooling Challenge
- Build and Test a Pooling Step
- Pooling Quiz
- Explain Pooling
- Pooling Vocabulary
- When Pooling Hurts
- Chapter Complete!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
CNN Architectures29 lessons· meets 20 requirements inADEKCBSECISCE + statesIB+5
Lessons
- The CNN Hall of Fame
- LeNet-5: The Pioneer
- Inside LeNet
- Shape Tracker
- The 14-Year Gap
- AlexNet Breakthrough
- Why 2012?
- Innovation Match
- Which Bottleneck?
- VGGNet: Keep It Simple
- Why 3×3 Wins
- Count the Weights
- The Depth Problem
- GoogLeNet Inception
- CNN Timeline
- ResNet Revolution
- Model Size Lab
- Architecture Choice in Practice
- Skip Connection Magic
- Residual Arithmetic
- Architecture Match
- Modern Architectures
- Transfer Learning
- Transfer or Scratch?
- Choosing the Right CNN
- Architecture Quiz
- Explain CNN Architectures
- Architecture Vocabulary
- Chapter Complete!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- ADEKFootprint / resource cost of AI
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Evaluate licenses that limit or restrict use of computational artifacts when using resources such as libraries.
- CSTA (US)Explain the beneficial and harmful effects that intellectual property laws can have on innovation.
- CSTA (US)Design ways to minimize negative environmental impacts of AI and communicate those ways to others.
- CSTA (US)Students examine security implications and environmental impacts of AI systems within Systems & Security.
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how design and deployment decisions affect an AI system's resource consumption (Engage 5, advanced).
- OECD–EC (Europe)Consider the ethical consequences of AI systems throughout their life-cycle, including environmental impact and societal impact such as platformisation of work and low-cost labelling labour (AI example 57, competence 4.4).
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Building a CNN30 lessons· meets 14 requirements inADEKCBSECISCE + statesIB+3
Lessons
- Time to Build!
- The CNN Recipe
- Keras Setup
- Sequential or Not?
- Create the Model
- Which Block Does What?
- Layer Parameters
- Trace the Shapes
- Add Classification Head
- Layer Order Game
- Builder Curiosity
- Compile the Model
- Compile Match
- Train with model.fit()
- CNN Build Lab
- Building Models for Real Apps
- Epochs & Batches
- Count the Updates
- Evaluate & Predict
- Complete CNN Code
- Tensor Shapes: Reshape and Broadcast
- Will It Broadcast?
- Predict the Shape
- Adapt the Recipe
- Workflow Order
- CNN Building Challenge
- Building CNN Quiz
- Explain Building a CNN
- CNN Building Terms
- Chapter Complete!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from demonstrating how training data are labeled by class in a tool like Teachable Machine (3-5), to explaining the difference between training a reasoning model and using it on new data (6-8), to illustrating all steps of building a classifier/predictor with machine learning, including choosing features, labeling, cross-validation, and test-set evaluation (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Data Augmentation27 lessons· meets 14 requirements inADEKCBSECISCE + statesIB+3
Lessons
- More variety from the same photos
- Why augmentation matters
- Does the label survive?
- Flip it!
- Flip or not?
- Tap to discover augmentation types
- ImageDataGenerator in action
- Read the ranges
- Color augmentations
- Match augmentation to use case!
- Augmentation Curiosity
- When NOT to augment
- Sort: Safe vs risky augmentations!
- Real-time augmentation
- Count the views
- Augmentation Mix Lab
- Augmentation in Real Products
- flow_from_directory
- Build an augmentation pipeline!
- Augmented copy or not?
- Keep validation real
- Training only, or both?
- Complete augmentation pipeline
- Data Augmentation Quiz!
- Explain Data Augmentation
- Master augmentation vocabulary
- Data Augmentation complete
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from describing what an intelligent agent must 'know' to make sense of a spoken question (K-2), to demonstrating how a text-to-speech/speech-recognition system resolves ambiguity using statistical context (3-5), to classifying an image scene type and describing the domain knowledge needed to understand it (6-8), to analyzing an online image dataset (e.g. ImageNet, Coco) to describe how statistical domain knowledge is derived for a computer vision system (9-12).
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Transfer Learning Introduction27 lessons· meets 15 requirements inADEKCBSECISCE + statesIB+3
Lessons
- Standing on the shoulders of giants!
- What is Transfer Learning?
- Transfer or Not?
- Why it works: Feature hierarchy
- Borrowed Vision Intelligence
- Tap to discover famous pre-trained models!
- Loading pre-trained models
- Freezing layers
- Adding a custom head
- Count the Head
- Match the transfer learning terms!
- Frozen, Trained or No Weights?
- Two strategies
- Sort: Feature Extraction vs Fine-Tuning!
- Fine-tuning in code
- Fine-Tune Debugging
- Transfer Strategy Lab
- Transfer Learning in Practice
- Order the transfer learning steps!
- When to use transfer learning
- Where would transfer learning help?
- Complete transfer learning example
- Read the Pipeline
- Transfer Learning Quiz!
- Explain Transfer Learning
- Master transfer learning terms
- Transfer Learning Expert!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from describing what an intelligent agent must 'know' to make sense of a spoken question (K-2), to demonstrating how a text-to-speech/speech-recognition system resolves ambiguity using statistical context (3-5), to classifying an image scene type and describing the domain knowledge needed to understand it (6-8), to analyzing an online image dataset (e.g. ImageNet, Coco) to describe how statistical domain knowledge is derived for a computer vision system (9-12).
- AI4K12 (US)Progress from illustrating why training a classifier for a broad concept like 'dog' requires large, diverse image data by browsing datasets such as ImageNet or Coco (3-5), to illustrating how objects in images are segmented and labeled with bounding boxes to build an object-recognition training set (6-8), to evaluating a real AI training dataset's size, acquisition/labeling method, storage cost, and time-to-produce, including where bias can enter (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Famous CNN Models27 lessons· meets 13 requirements inADEKCBSECISCE + statesIB+3
Lessons
- The models that changed everything!
- LeNet-5: Where it all began
- Read the Scoreboard
- AlexNet: The game changer
- AlexNet's Ingredients
- Tap to reveal CNN milestones!
- VGG: Simple but powerful
- Stacking Small Filters
- GoogLeNet: many filter sizes at once
- Match CNN to innovation!
- Squeeze Before You Convolve
- ResNet: going much deeper
- Run a Residual Block
- Sort models by layer count!
- Modern efficient models
- Count a Separable Conv
- Model File Size Lab
- Choosing Models for Products
- Tap to compare parameter counts!
- Order the CNN evolution!
- Pick the right model!
- Using pre-trained models
- Choosing the right model
- Famous CNN Models Quiz!
- Explain Famous CNN Models
- Master CNN model vocabulary
- CNN Models Expert!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Object Detection Basics27 lessons· meets 16 requirements inADEKCBSECISCE + statesIB+3
Lessons
- Finding objects in images!
- Three computer vision tasks
- Which Vision Task?
- Bounding boxes explained
- Read the Box
- Tap to discover detection methods!
- R-CNN: Region-based detection
- YOLO: Real-time detection
- Match detector to feature!
- The Frame Budget
- IoU: Intersection over Union
- Calculate the IoU
- Non-Max Suppression (NMS)
- Kept or Removed?
- Sort: Two-stage vs One-stage!
- Frame Budget Lab
- Object Detection in Products
- Tap to see detection applications!
- Order the detection steps!
- Which needs object detection?
- YOLO in practice
- Read the Detector Output
- mAP: Mean Average Precision
- Object Detection Quiz!
- Explain Object Detection
- Master detection vocabulary
- Object Detection Expert!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEUnderstand and visualize the computer's ability to identify alphabets and handwriting; hands-on Computer Vision unit content (Unit 5: Computer Vision, 20 practical hours).
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesGive at least one real-world application example of robots in domestic, industry, medical, defence, entertainment and agriculture domains
- CISCE + statesExplain New Age Robotic Systems (warehouse, assistant, smart-home/school/mobility robots, autonomous cars/drones, medical/healthcare robotics) and why NARS are becoming increasingly relevant and possible
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Face Recognition27 lessons· meets 21 requirements inADEKCBSECISCE + statesIB+6
Lessons
- Your face is your password!
- Detection vs Recognition
- Detect, Verify or Identify?
- Face detection methods
- A Face as Numbers
- Tap to discover face landmarks!
- Face embeddings: The magic!
- Comparing face embeddings
- Same Person or Not?
- Match term to meaning!
- One-to-One or One-to-Many?
- How face models learn
- Triplet Check
- Sort: Detection vs Recognition!
- Tap to see face recognition uses!
- Embedding Similarity Lab
- Face Recognition in Devices
- Face recognition pipeline
- Order the face recognition steps!
- Privacy concerns
- Ethical or concerning?
- Preventing spoofing
- False Accept or False Reject?
- Face Recognition Quiz!
- Explain Face Recognition
- Master face recognition terms
- Face Recognition Expert!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEUnderstand and visualize the computer's ability to identify alphabets and handwriting; hands-on Computer Vision unit content (Unit 5: Computer Vision, 20 practical hours).
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
- ChinaStudents build a three-dimensional 'technical principle → system architecture → social impact' way of reasoning about AI
Real-World CV Applications29 lessons· meets 22 requirements inADEKCBSECISCE + statesIB+6
Lessons
- CV is changing the world!
- Healthcare: Saving lives
- The Accuracy Trap
- Agriculture: Feeding the world
- Which CV Job?
- Segmentation Inside: U-Net
- U-Net Parts
- Score the Mask
- Tap to discover CV applications!
- Transportation: Safer roads
- Beat the Frame Budget
- Retail: Shopping reimagined
- Match CV application to industry!
- CV Applications Curiosity
- Manufacturing: Zero defects
- Move the Threshold
- Smart cities: Urban AI
- Sort: Object Detection vs Classification!
- CV Deployment Lab
- A Real Product Example
- Accessibility: CV for all
- Order a medical imaging pipeline!
- Which uses computer vision?
- Real-world challenges
- Edge or Cloud?
- Real-World CV Applications Quiz!
- Explain CV Applications
- CV applications vocabulary
- CV Applications Expert!
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEIdentify and appreciate AI and describe its applications in daily life; relate, apply and reflect on Human-Machine Interaction across the three domains of AI - Data, Computer Vision, and Natural Language Processing; undergo assessment for AI-readiness skills (Unit 1: AI Reflection, Project Cycle and Ethics).
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesGive at least one real-world application example of robots in domestic, industry, medical, defence, entertainment and agriculture domains
- CISCE + statesIdentify applications and benefits of AI across commercial, industry, medical/healthcare, defence, banking, entertainment, transport, security and agriculture domains
- CISCE + statesExplain New Age Robotic Systems (warehouse, assistant, smart-home/school/mobility robots, autonomous cars/drones, medical/healthcare robotics) and why NARS are becoming increasingly relevant and possible
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Evaluate and refine computational artifacts to make them more usable and accessible.
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from identifying current uses of AI and how they have impacted society (K-2), to identifying how sectors of society such as manufacturing, retail, and healthcare are changing due to AI (3-5), to comparing AI-driven societal change to previous industrial revolutions (6-8), to predicting how a specific sector of society is likely to change in the short and intermediate term as a result of AI, referencing Amara's Law (9-12).
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- SingaporeGive examples of common personal and business tasks that can be performed well by AI: face recognition, voice recognition, image classification and spam filtering.
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
- Korea, Japan, HKUnderstand and apply the concepts and methods of information design (including universal design, usability and accessibility) needed for effective communication with a given audience, and evaluate/improve created content.
- Australia + NZDesign the user experience of a digital system generating or processing data, and evaluate the suitability of user interfaces for defined purposes and end-users, applying accessibility standards for diverse audiences
Unit 3 Review - Computer Vision Mastery25 lessons· meets 15 requirements inADEKCBSECISCE + statesIB+3
Lessons
- Unit 3 review: pixels to faces
- Review: Images as data
- Why Did The AI Slip Up?
- Pixels to numbers
- Review: CNN layers
- Trace the shapes
- Match CNN layer to purpose!
- Tiny Timing, Big Impact
- Tap to review famous CNNs!
- Review: Transfer learning
- Review: Data augmentation
- Sort: Safe vs risky augmentations!
- Review: Object detection
- Boxes and overlap
- Review: Face recognition
- Match or no match?
- Tune A Vision System
- Why Phone Photos Feel Smart
- Order the face recognition steps!
- CNN or not?
- Unit 3 Comprehensive Quiz - Part 1!
- Unit 3 Comprehensive Quiz - Part 2!
- Explain The Vision Pipeline
- Unit 3 key vocabulary
- Unit 3 review complete
Requirements this topic helps meet
- ADEKFacial recognition: privacy/safety/fairness tension
- CBSEIdentify and appreciate AI and describe its applications in daily life; relate, apply and reflect on Human-Machine Interaction across the three domains of AI - Data, Computer Vision, and Natural Language Processing; undergo assessment for AI-readiness skills (Unit 1: AI Reflection, Project Cycle and Ethics).
- CBSEExplain computer vision and its significance in visual data analysis; understand key stages (acquisition, preprocessing, feature extraction, analysis); identify real-world applications in healthcare, surveillance, autonomous vehicles; analyse challenges (ethics, privacy, technical limitations); develop basic skills using OpenCV and deploying ML models online (Unit 3).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- CISCE + statesIntroduce Computer Vision and OpenCV applications in real-world scenarios: image classification, object detection, facial recognition, colour detection and object tracking (brief explanation with examples)
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDescribe deep learning at an introductory level as a machine learning technique
- AI4K12 (US)Progress from demonstrating figure/ground segmentation in an image (K-2), to illustrating how occlusion changes the outlines of objects (3-5), to describing how edge detectors compose into more complex feature detectors such as letter/shape detectors (6-8), to demonstrating how perceptual reasoning at a higher abstraction level (objects, surfaces, boundaries) draws on lower levels (contours, edges, pixels) (9-12).
- AI4K12 (US)Progress from describing what an intelligent agent must 'know' to make sense of a spoken question (K-2), to demonstrating how a text-to-speech/speech-recognition system resolves ambiguity using statistical context (3-5), to classifying an image scene type and describing the domain knowledge needed to understand it (6-8), to analyzing an online image dataset (e.g. ImageNet, Coco) to describe how statistical domain knowledge is derived for a computer vision system (9-12).
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Unit 4: Natural Language NLP 12 topics
Deep Learning for Text26 lessons· meets 9 requirements inCBSECISCE + statesIBCSTA (US)+3
Lessons
- story
- concept
- Why Did The Meaning Flip?
- concept
- Predict the IDs
- interactive.tapReveal
- concept
- Close in Meaning?
- concept
- concept
- Trace the Memory
- game.matching
- Does Order Matter?
- concept
- game.sorting
- concept
- Read the Model
- Tune A Text Model
- Why Smart Replies Feel Fast
- game.dragDrop
- activity.spotTheAI
- concept
- quiz
- Explain Sequence Understanding
- vocabulary
- reward
Requirements this topic helps meet
- CBSEIdentify and appreciate AI and describe its applications in daily life; relate, apply and reflect on Human-Machine Interaction across the three domains of AI - Data, Computer Vision, and Natural Language Processing; undergo assessment for AI-readiness skills (Unit 1: AI Reflection, Project Cycle and Ethics).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Word Embeddings31 lessons· meets 14 requirements inCBSECISCE + statesIBCSTA (US)+3
Lessons
- Words as Numbers
- The One-Hot Problem
- How Did Search Find That?
- Dense Embeddings
- Meaning Has Shape
- Encoding Match-Up
- Keyword or Meaning?
- Word Arithmetic
- Relationship Sorting
- Fact or Stereotype?
- Measuring Similarity
- Compute the Cosine
- Choosing Dimensions
- PCA: Squash Vectors, Keep the Spread
- PCA Step Match
- Run PCA
- Learning Embeddings
- Keras Embedding Layer
- Check the Layer
- Tune An Embedding Space
- Why Search Understands Synonyms
- Pre-trained Embeddings
- Embeddings Everywhere
- Applications Match
- Spot Embedding Apps
- Compare Word Vectors in Code
- Embeddings Quiz
- Explain Word Vectors
- Embedding Vocabulary
- Key Takeaways
- Chapter Complete!
Requirements this topic helps meet
- CBSEAnalyse data (numbers/images), find patterns/relations, and understand the role of Math (statistics, linear algebra, probability, calculus) in AI; understand statistics and probability concepts and apply them to real-life scenarios such as disaster management, sports, disease prediction, weather forecast, traffic estimation (Unit 3: Math for AI).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from describing the sounds that make up spoken language and giving example words for each vowel sound (K-2), to illustrating how sequences of sounds are recognized as candidate words despite unclear signals (3-5), to illustrating how sequences of words are recognized as phrases by resolving ambiguity with higher-level information (6-8), to illustrating the full abstraction hierarchy from waveform to sentence in speech understanding (9-12).
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- AI4K12 (US)Progress from demonstrating knowledge of language structure by generating plausible/implausible novel words or unscrambling sentences (K-2), to demonstrating that human language is infinite via sentence extension (3-5), to using a parser to display a sentence's syntactic (parse-tree) structure and explaining the node categories (6-8), to demonstrating how a small context-free grammar can parse or generate simple sentences, and its limits with agreement constraints (9-12).
- AI4K12 (US)Progress from demonstrating how a computer produces different verb tense/agreement forms (K-2), to experimenting with a speech-to-text system's ability to resolve ambiguous words via context (3-5), to illustrating how word embeddings support reasoning about word meaning, e.g. outlier detection and analogy completion (6-8), to demonstrating how a small context-free grammar parses or generates simple sentences and where it falls short, e.g. subject-verb agreement (9-12).
- OECD–EC (Europe)Weigh the benefits and disadvantages of using AI-driven search engines, including privacy and commercial-interest trade-offs (AI example 31, competence 1.1).
- OECD–EC (Europe)Be aware that AI algorithms may embody a commercial or political message rather than only serving the user's information need, with possible negative consequences such as reproducing stereotypes or spreading misinformation (AI example 48, competence 1.2).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Word2Vec Basics30 lessons· meets 10 requirements inCISCE + statesIBCSTA (US)AI4K12 (US)+2
Lessons
- The Algorithm Revolution
- CBOW Explained
- How Did It Learn Espresso?
- Skip-gram Explained
- CBOW or Skip-gram?
- Method Matching
- Billions Of Words, One Model
- Context Window
- Count the Pairs
- Training Process
- Negative Sampling
- Training Examples
- Gensim Library
- Pre-trained Word2Vec
- Predict the Output
- Drawing the Space: t-SNE and UMAP
- Honest Map Reading
- Run the Neighbour Odds
- Tuning Word2Vec
- Tune Word2Vec Training
- Why News Feeds Group Topics
- Parameter Match
- Word2Vec Applications
- Where Word2Vec Falls Short
- Spot Word2Vec Apps
- Word2Vec Quiz
- Explain Context Learning
- Word2Vec Vocabulary
- Key Takeaways
- Chapter Complete!
Requirements this topic helps meet
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- AI4K12 (US)Progress from demonstrating how a computer produces different verb tense/agreement forms (K-2), to experimenting with a speech-to-text system's ability to resolve ambiguous words via context (3-5), to illustrating how word embeddings support reasoning about word meaning, e.g. outlier detection and analogy completion (6-8), to demonstrating how a small context-free grammar parses or generates simple sentences and where it falls short, e.g. subject-verb agreement (9-12).
- AI4K12 (US)Progress from giving commonsense explanations of human behavior useful for a robot to know, e.g. choosing a plausible story ending (K-2), to explaining what everyday cultural and naive-physics knowledge a computer needs to understand a story (3-5), to explaining the naive-physics inference rules (mass, volume, conservation, forces) needed to solve a physics reasoning problem (6-8), to explaining the cultural, naive-physics, and folk-psychology knowledge needed to interpret a fable, fairytale, or Winograd-schema sentence (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Recurrent Neural Networks27 lessons· meets 9 requirements inCISCE + statesIBCSTA (US)AI4K12 (US)+2
Lessons
- The Memory Problem
- Sequence Data Types
- How Did The Meaning Stay Alive?
- The Hidden State
- Trace the Hidden State
- RNN vs Regular NN
- Your Keyboard Uses Memory
- Vanishing Gradient
- Shrinking Signal
- LSTM: The Solution
- LSTM Information Flow
- GRU: Simplified LSTM
- Gate Matching
- Bidirectional RNNs
- Which Sequence Model?
- Tune Sequence Memory
- Why Predictive Text Feels Personal
- RNN Applications
- Building RNNs in Keras
- Read the Code
- Layer Matching
- Spot RNN Applications
- RNN Quiz
- Explain Hidden State Memory
- RNN Vocabulary
- Key Takeaways
- Chapter Complete!
Requirements this topic helps meet
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from describing the sounds that make up spoken language and giving example words for each vowel sound (K-2), to illustrating how sequences of sounds are recognized as candidate words despite unclear signals (3-5), to illustrating how sequences of words are recognized as phrases by resolving ambiguity with higher-level information (6-8), to illustrating the full abstraction hierarchy from waveform to sentence in speech understanding (9-12).
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
LSTM Networks27 lessons· meets 8 requirements inCISCE + statesIBCSTA (US)AI4K12 (US)+2
Lessons
- The Memory Challenge
- The Cell State Highway
- Why Didn't The Model Forget?
- The Three Gates
- A Memory Highway Inside AI
- Forget Gate Details
- Input Gate Details
- Gate Matching Game
- Autocomplete Depends On Memory
- Updating Cell State
- Update the Cell by Hand
- Output Gate Details
- Hidden, Not Forgotten
- LSTM Flow Order
- Gate Function Sorting
- Tune LSTM Memory
- Why Voice Dictation Feels Coherent
- Building LSTM in Keras
- Count the Weights
- Stacked LSTMs
- LSTM Applications
- Spot LSTM Applications
- LSTM Quiz
- Explain LSTM Memory
- LSTM Vocabulary
- Key Takeaways
- Chapter Complete!
Requirements this topic helps meet
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Text Classification26 lessons· meets 8 requirements inCISCE + statesIBCSTA (US)AI4K12 (US)+2
Lessons
- Sorting Text Into Boxes
- Binary vs Multi-Class
- Why Did The Email Get Flagged?
- Sentiment Analysis
- Sorting Meaning At Scale
- Spam Detection
- Classification Sorting
- Which Kind of Classification?
- Text Preprocessing
- Preprocessing Order
- Predict the Bag of Words
- LSTM for Classification
- Read the Output Layer
- Tune A Classifier
- Why Your Inbox Feels Organized
- Layer Matching
- Evaluation Metrics
- Compute the Metrics
- Applications Everywhere
- Applications Match
- Spot Classification Apps
- Classification Quiz
- Explain Text Classification
- Classification Vocabulary
- Key Takeaways
- Chapter Complete!
Requirements this topic helps meet
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Sequence to Sequence Models25 lessons· meets 8 requirements inCISCE + statesIBCSTA (US)AI4K12 (US)+2
Lessons
- The Translation Challenge
- Encoder and Decoder
- Why Can't One Vector Hold It All?
- Inside the Encoder
- One Sequence Becomes Another
- The Context Vector
- Encoder Parts Match
- Fixed Size, Any Length
- Inside the Decoder
- Trace a Decode
- The Telephone Problem
- Seq2Seq Step Order
- Teacher Forcing
- Teacher Forcing in Action
- Tune Seq2Seq Capacity
- Why Translation Apps Need Structure
- Training vs Testing
- Seq2Seq Applications
- Spot Seq2Seq
- Keras Seq2Seq Code
- Read the Encoder Code
- Seq2Seq Quiz
- Explain Encoder And Decoder
- Key Terms
- Seq2Seq Master!
Requirements this topic helps meet
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Attention Mechanism Introduction26 lessons· meets 8 requirements inCISCE + statesIBCSTA (US)AI4K12 (US)+2
Lessons
- The Forgetful Translator
- The Bottleneck Problem
- How Did The Decoder Find The Right Word?
- What is Attention?
- Bottleneck Check
- The Spotlight Analogy
- Query, Key, Value
- QKV Match Up
- Query, Key or Value?
- Computing Attention
- Compute the Weights
- Attention Steps Order
- Attention in Translation
- Attention Heatmaps
- Read the Heatmap
- Softmax Spotlight
- Why Translation Got Better
- Before and After Attention
- Types of Attention
- Attention in Code
- Spot Attention
- Change the Query, Watch Attention
- Attention Quiz
- Explain Attention
- Key Terms
- Attention Expert!
Requirements this topic helps meet
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Transformer Preview27 lessons· meets 12 requirements inADEKCBSECISCE + statesIB+5
Lessons
- The AI Revolution
- Why Transformers?
- Count the Steps
- Transformer Architecture
- Attention Became The Whole Engine
- Self-Attention Explained
- RNN vs Transformer
- Weigh the Words
- Multi-Head Attention
- Multi-Head in Action
- Heads by the Numbers
- The Position Problem
- Positional Encoding
- Order Blind?
- Transformer Layer Order
- Tune A Transformer
- Why Chat Assistants Feel Different
- Encoder vs Decoder
- Encoder or Decoder?
- Famous Transformers
- Pick the Transformer Shape
- The ChatGPT Connection
- Spot Transformer Apps
- Knowledge Check
- Explain Transformers
- Key Terms
- Transformer Pioneer!
Requirements this topic helps meet
- ADEKClassification vs generation, how generative AI is trained
- CBSEDefine Generative AI and classify different kinds; explain how Generative AI works and how it learns, differentiate from conventional AI; apply Generative AI tools to create content; understand ethical considerations of using Generative AI (Unit 4: Introduction to Generative AI).
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- AI4K12 (US)Progress from identifying devices in daily life that use AI, e.g. smartphones, smart assistants (K-2), to describing how AI-powered services (search, recommenders, autonomous vehicles) are used in daily life (3-5), to examining an aspect of daily life predicted to change due to AI, e.g. transportation and self-driving cars (6-8), to explaining the kinds of cultural debates that arise as AI evolves, e.g. AI-generated art, LLM use in homework, data rights (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
- ChinaStudents understand the technical characteristics of generative AI and its significance to national strategy
Multilingual NLP27 lessons· meets 10 requirements inCISCE + statesIBCSTA (US)AI4K12 (US)+2
Lessons
- The Language Challenge
- High vs Low Resource
- Speakers or Data?
- Multilingual Challenges
- Which Challenge Is It?
- Language Features
- How Far Can One Model Reach?
- One Model, Many Languages
- Cross-Lingual Transfer
- Near in Meaning Space
- Multilingual Models
- High vs Low Resource
- Script Challenges
- Balance a Multilingual Model
- Why Translation Apps Need More Than Dictionaries
- Tokenization Examples
- The Token Tax
- Preserving Languages
- Multilingual Techniques
- Translation Revolution
- Pick the Right Fix
- Multilingual AI in Action
- Future of Multilingual AI
- Knowledge Check
- Explain Cross-Lingual Transfer
- Key Terms
- Multilingual NLP Complete
Requirements this topic helps meet
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- AI4K12 (US)Progress from demonstrating knowledge of language structure by generating plausible/implausible novel words or unscrambling sentences (K-2), to demonstrating that human language is infinite via sentence extension (3-5), to using a parser to display a sentence's syntactic (parse-tree) structure and explaining the node categories (6-8), to demonstrating how a small context-free grammar can parse or generate simple sentences, and its limits with agreement constraints (9-12).
- AI4K12 (US)Progress from giving commonsense explanations of human behavior useful for a robot to know, e.g. choosing a plausible story ending (K-2), to explaining what everyday cultural and naive-physics knowledge a computer needs to understand a story (3-5), to explaining the naive-physics inference rules (mass, volume, conservation, forces) needed to solve a physics reasoning problem (6-8), to explaining the cultural, naive-physics, and folk-psychology knowledge needed to interpret a fable, fairytale, or Winograd-schema sentence (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Chatbot with Deep Learning26 lessons· meets 10 requirements inCBSECISCE + statesIBCSTA (US)+3
Lessons
- The Breakthrough
- Neural Networks Explained
- Predict the Neuron
- The Three Layers
- Layers Build Abstraction
- What Is Deep Learning?
- Network Architectures
- Architecture Match
- Which Architecture?
- Traditional ML vs Deep Learning
- Sort ML Types
- Activation Functions
- Run the Activations
- Tune a Chatbot Network
- Why Modern Chatbots Feel Context-Aware
- How Networks Learn
- Training Steps
- One Training Step
- Forward or Backward?
- Chatbots and Deep Learning
- Pick the Next Token
- Spot Deep Learning
- Knowledge Check
- Explain Deep Learning for Chatbots
- Key Terms
- Deep Learning Expert!
Requirements this topic helps meet
- CBSEDifferentiate AI, ML, and DL and common data terminologies; familiarise with supervised, unsupervised, and reinforcement learning-based approaches; understand subcategories (classification, regression, clustering, association, ANN, CNN); understand neural networks and how AI makes a decision (Unit 2: Advanced Concepts of Modeling in AI).
- CISCE + statesExplain the brief history and primary elements of AI including Machine Learning (ML) vs Deep Learning (DL)/neural networks, and applications of ML/DL such as image recognition, computer vision, speech recognition and information retrieval
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Unit 4 Review24 lessons· meets 11 requirements inCBSECISCE + statesIBCSTA (US)+3
Lessons
- Your NLP Journey
- NLP Foundations Review
- Count the Bag
- One Model, So Many Jobs?
- Embeddings Recap
- Close or Far?
- One Toolkit, Many Tasks
- NLP Concepts Match
- Why One Backbone Matters
- Sequence Models Review
- Attention Recap
- Attention by Hand
- Transformers Review
- Model Architecture Sort
- Multilingual Review
- Review the NLP Stack
- One NLP Stack, Many Writing Features
- Deep Learning Match
- NLP Evolution Order
- Spot NLP in Action
- Unit 4 Final Quiz
- Explain the Big Unit 4 Idea
- Unit 4 Vocabulary
- NLP Master!
Requirements this topic helps meet
- CBSEStudy Natural Language Processing concepts and applications as Unit 6 of the Class X curriculum (20 theory + 7 practical hours); exact sub-unit learning outcomes not extracted from the source PDF pages read.
- CISCE + statesExplain Deep Learning and demonstrate image/pattern recognition using a Convolutional Neural Network (CNN) example (e.g. classifying images of cats, dogs, birds)
- IBDescribe deep learning at an introductory level as a machine learning technique
- CSTA (US)Describe how current AI models (e.g., LLMs) use data representation.
- CSTA (US)Choose and use an appropriate representation of complex data for processing by an AI algorithm.
- AI4K12 (US)Progress from discussing why intelligent agents need to understand languages other than English (K-2), to discussing why domain knowledge must be broad enough to cover accents, dialects, and musical genres (3-5), to describing how a vision system can exhibit cultural bias if it lacks knowledge of objects outside its creators' culture, e.g. non-US traffic signs (6-8), to describing technical difficulties (e.g. low-contrast faces, children's speech) in making perception systems work well for diverse groups (9-12).
- AI4K12 (US)Progress from identifying the features that make objects in a collection unique and organizing them into a feature table (K-2), to constructing a feature-vector representation for a set of objects and showing that similar objects lie close together in feature space (3-5), to explaining how word embeddings represent words as numeric vectors for tasks like machine translation (6-8), to describing how a transformer network maps sequences of word-embedding vectors to output sequences (9-12).
- AI4K12 (US)Progress from demonstrating what an intelligent assistant like Siri or Alexa can and cannot do (K-2), to demonstrating question types a search engine or assistant can and cannot answer (3-5), to describing NLP tasks such as summarization, sentiment analysis, question answering, and machine translation (6-8), to describing NLP approaches ranging from keyword/template matching to deep neural networks, and when each is appropriate (9-12).
- AI4K12 (US)Progress from giving commonsense explanations of human behavior useful for a robot to know, e.g. choosing a plausible story ending (K-2), to explaining what everyday cultural and naive-physics knowledge a computer needs to understand a story (3-5), to explaining the naive-physics inference rules (mass, volume, conservation, forces) needed to solve a physics reasoning problem (6-8), to explaining the cultural, naive-physics, and folk-psychology knowledge needed to interpret a fable, fairytale, or Winograd-schema sentence (9-12).
- OECD–EC (Europe)Be aware that AI systems are typically developed in English-speaking contexts and can work less accurately in non-English contexts, e.g. machine translation between less-used language pairs (AI example 55, competence 2.5).
- Korea, Japan, HKApply deep-learning-based computer vision, speech recognition and natural-language-processing technologies, choosing appropriate deep-learning techniques to solve real-life and cross-disciplinary problems.
Unit 5: AI Ethics Deep Dive 15 topics
Why AI Ethics Matters22 lessons· meets 21 requirements inADEKCISCE + statesCSTA (US)AI4K12 (US)+4
Lessons
- The Robot's Dilemma
- What is AI Ethics?
- The Unseen Rule
- Four Pillars of AI Ethics
- Invisible Rules, Real Impact
- Pillar Match Game
- Why Explanations Matter
- AI Decisions in Real Life
- Sort the Stakes!
- When AI Goes Wrong
- Why AI Makes Mistakes
- Diagnose the Failure
- Run the Four-Fifths Check
- Tune an Ethics Score
- Why Recommendation Apps Need Ethics
- EU AI Act Risk Tiers
- Steps to Ethical AI
- Spot the Ethics Issue!
- Ethics Knowledge Quiz
- Explain Human Review
- Ethics Vocabulary
- Ethics Champion!
Requirements this topic helps meet
- ADEKDraft/commit to ethical-use guidelines
- CISCE + statesDescribe ethical considerations in AI including bias, prejudice, fairness, accountability, transparency, interpretability and explainability
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from discussing what makes a system fair vs. unfair through a fair-distribution exercise (K-2), to evaluating whether an AI system meets the design criteria of transparency and explainability (3-5), to evaluating whether an AI system meets the design criteria of accountability and respect for privacy (6-8), to analyzing whether an AI system satisfies the full set of ethical design criteria: fairness, transparency, explainability, accountability, respect for privacy, and adherence to societal values (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how AI shapes beliefs and behaviours (Engage 1, advanced level).
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Design improvements to an AI system based on technical constraints, ethics, and user needs to promote societal benefit (Shape 4, advanced).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)Recognise that AI applications directly interacting with humans and making decisions about their lives (e.g. CV-sorting, exam scoring) can be controversial even where other AI applications are not (AI example 63, competence 2.3).
- UNESCOUnderstand that human accountability is a legal obligation of AI creators and AI service providers, who are answerable for safety incidents, ethical risks in design/training, and misuses of their AI service (CG4.2.1.1).
- UNESCOEvaluate the consequences of institutional use of AI to make high-stakes decisions about people (e.g. profiling students, hiring, loan/assessment scoring) and understand why human accountability must not be ceded to AI in such decisions (CG4.2.1.2).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- UNESCOApply an integral, ethics-by-design set of principles across the whole AI life cycle, e.g. simulating the role of a 'chief ethics officer' drafting a checklist of ethical criteria to audit each step of an AI system's design (CG4.3.2.1).
- SingaporeExplain how unethical use of AI or using AI with biased data can lead to negative consequences.
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
Bias in AI Systems22 lessons· meets 17 requirements inADEKCSTA (US)AI4K12 (US)OECD–EC (Europe)+3
Lessons
- The Interview That Never Was
- What is AI Bias?
- The Resume Riddle
- Three Types of AI Bias
- Small Bias, Big Loop
- Match the Bias Type!
- Bias Hides in Proxy Features
- Where AI Bias Hurts Most
- Sort the Bias Impact!
- Famous AI Bias Cases
- Diagnose the Case
- How to Fight AI Bias
- Run the Four-Fifths Check
- Measure a Fairness Score
- Why Diverse Data Matters in Vision AI
- Match the Fix
- Build Fair AI Steps
- Spot the Bias!
- Bias Expert Quiz
- Explain How Bias Slips In
- Bias Vocabulary
- Bias Fighter!
Requirements this topic helps meet
- ADEKExamine dataset bias, propose fixes
- ADEKDraft/commit to ethical-use guidelines
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from examining a labeled dataset to identify problems that could cause incorrect predictions, e.g. an all-green 'healthy foods' set (K-2), to examining training-data features/labels for potential sources of bias such as Amazon's biased resume sorter (3-5), to explaining how unbalanced training data biases a classifier, e.g. faces or loan applications (6-8), to investigating demographic imbalances in training data using data-visualization tools such as histograms or Pandas (9-12).
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Weigh the benefits and disadvantages of using AI-driven search engines, including privacy and commercial-interest trade-offs (AI example 31, competence 1.1).
- OECD–EC (Europe)Be aware that AI algorithms may embody a commercial or political message rather than only serving the user's information need, with possible negative consequences such as reproducing stereotypes or spreading misinformation (AI example 48, competence 1.2).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- SingaporeExplain how unethical use of AI or using AI with biased data can lead to negative consequences.
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
Fairness in Machine Learning27 lessons· meets 22 requirements inADEKCBSECISCE + statesEngland + Cambridge+6
Lessons
- The Unfair Algorithm
- Defining Fairness
- The Same Score, Different Outcome
- Measuring Fairness
- Run the Checks
- Fairness Has More Than One Meaning
- Match Fairness Types!
- Why One Metric Is Never Enough
- The Fairness Trade-Off
- Sort by Situation!
- Techniques for Fair AI
- Before, During or After?
- Fairness in the Real World
- Tune a Fairness Strategy
- Why Lenders Audit Approval Models
- Audit It in Python
- Audit a Shortlist in Python
- Predict the Shortlist Check
- A Signal, Not a Verdict
- Write and Run Your Own Audit
- Complete the Code, Read the Output
- Build Fair AI Steps
- Spot Fairness Issues!
- Fairness Knowledge Quiz
- Explain the Fairness Trade-Off
- Fairness Vocabulary
- Fairness Champion!
Requirements this topic helps meet
- ADEKDraft/commit to ethical-use guidelines
- ADEKPython analysis of a people-rating/ranking dataset
- CBSEAcquire introductory Python programming skills: variables, arithmetic/logical/comparison/assignment operators, data types, print()/input(), conditional and iterative statements (if/for/while), and lists, with hands-on practical exercises (Unit 5: Introduction to Python; Part C practical exam).
- CISCE + statesWrite Python programs using variables, data types, operators, conditional statements (if/elif/else), loops (for/while, nested, break/continue/pass) and user-defined functions
- England + CambridgeDescribe the operation of a compiler and an interpreter, including how high-level language is translated and how errors are reported, and explain the role of an IDE and its common functions
- CSTA (US)Decompose problems into smaller components through systematic analysis, using constructs such as procedures, modules, and/or objects.
- CSTA (US)Create artifacts by using procedures within a program, combinations of data and procedures, or independent but interrelated programs.
- CSTA (US)Test and refine computational artifacts to reduce bias and equity deficits.
- CSTA (US)Demonstrate ways a given algorithm applies to problems across disciplines.
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- Korea, Japan, HKUnderstand means of expressing algorithms and be able to write programs (including defining/using functions) to make use of computers and networks, and evaluate/improve the resulting program.
- Australia + NZRecognise how the identification and mitigation of bias is a critical aspect of machine learning, and critically evaluate the reliability, validity and ethics of AI-generated outputs and predictive strategies
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
Privacy and AI22 lessons· meets 28 requirements inADEKEngland + CambridgeCSTA (US)AI4K12 (US)+5
Lessons
- The Listening Speaker
- Data AI Collects
- The Sneaker Ad Surprise
- Privacy Risks with AI
- Tiny Signals Become Big Predictions
- Match Privacy Risks!
- Why 'Anonymous' Data Still Leaks
- Could This Identify You?
- Privacy-Preserving AI
- Do the Coin-Flip Maths
- Sort Privacy Methods!
- Privacy Laws Worldwide
- Which Right Applies?
- Privacy Protection Tips
- Privacy vs Accuracy Dial
- Why Phones Use Local Learning
- One Round of Federated Learning
- Spot Privacy Issues!
- Privacy Expert Quiz
- Explain Why Data Minimization Matters
- Privacy Vocabulary
- Privacy Guardian!
Requirements this topic helps meet
- ADEKDraft/commit to ethical-use guidelines
- England + CambridgeUnderstand how changes in technology affect safety, including new ways to protect online privacy and identity, and how to identify and report a range of concerns
- CSTA (US)Explain the privacy concerns related to the collection and generation of data through automated processes that may not be evident to users.
- CSTA (US)Evaluate the social and economic implications of privacy in the context of safety, law, or ethics.
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how AI shapes beliefs and behaviours (Engage 1, advanced level).
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Weigh the benefits and disadvantages of using AI-driven search engines, including privacy and commercial-interest trade-offs (AI example 31, competence 1.1).
- OECD–EC (Europe)Be aware that AI algorithms may embody a commercial or political message rather than only serving the user's information need, with possible negative consequences such as reproducing stereotypes or spreading misinformation (AI example 48, competence 1.2).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)Be aware that content shared publicly online can be used to train AI systems (e.g. facial-recognition training on shared photos), which may raise privacy concerns (AI example 45, competence 2.2).
- OECD–EC (Europe)Know that processing of personal data by AI (e.g. voice interactions with a virtual assistant) is subject to regulations such as the EU GDPR and can carry data protection, privacy and security risks (AI example 39, competence 4.2).
- UNESCOExplain that AI is human-led and that decisions by AI creators shape how AI systems affect human rights, human-AI interaction, and society; recognise that AI cannot replace human thinking or intellectual development (CG4.1.1.1).
- UNESCOCritically examine the dynamic relationship between human agency and machine agency, including cases where a degree of machine agency may be warranted (e.g. medical pattern detection) versus cases requiring human control (CG4.1.1.3).
- UNESCOCategorise general safety risks of AI and typical AI incidents, deepen knowledge of human rights to data protection and privacy and creators' legal responsibility to collect data only with informed consent, and practise strategies to ensure personal data is collected, used, shared, archived and deleted only with consent (CG4.2.2.3).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- SingaporeDevelop digital safety and security competencies as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital responsibility as one of the nine core NDLP digital competencies.
- Korea, Japan, HKUnderstand laws/regulations on information, the importance of information security, and individual responsibility and information morals in the information society.
- Australia + NZEvaluate how the Australian Privacy Principles apply to the storage and communication of data in digital systems, and assess/manage the risks of common cyberthreats (e.g. phishing, malware) to a system
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
- Australia + NZStudents understand that their privacy and personal data must be protected when using generative AI tools, and are taught not to input sensitive personal or others' data into such tools without appropriate safeguards
Explainable AI (XAI)24 lessons· meets 19 requirements inADEKCISCE + statesEngland + CambridgeCSTA (US)+5
Lessons
- The Mystery Rejection
- What is Explainable AI?
- The Silent No
- Black Box vs Interpretable
- Open the Box
- Rule or Neighbours?
- Follow the Tree
- Accuracy Is Not the Whole Story
- Match Model Types!
- Why Counterfactuals Feel Useful
- XAI Explanation Techniques
- Sort Explanation Types!
- XAI Tools Revealed
- Explain Tom's Score
- XAI in Real Life
- Tune an Explanation System
- Why Credit Apps Show Key Factors
- Pick the Right Explanation
- Build XAI Steps
- Spot XAI Issues!
- XAI Expert Quiz
- Explain Why Black Boxes Are Risky
- XAI Vocabulary
- XAI Expert!
Requirements this topic helps meet
- ADEKDraft/commit to ethical-use guidelines
- CISCE + statesDescribe ethical considerations in AI including bias, prejudice, fairness, accountability, transparency, interpretability and explainability
- England + CambridgeDescribe the main characteristics of AI: the collection of data and rules for using that data, the ability to reason, and (optionally) the ability to learn and adapt
- England + CambridgeExplain the basic operation and components of AI systems that simulate intelligent behaviour, limited to expert systems (knowledge base, rule base, inference engine, interface) and machine learning (a program's ability to automatically adapt its own processes and/or data)
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from discussing what makes a system fair vs. unfair through a fair-distribution exercise (K-2), to evaluating whether an AI system meets the design criteria of transparency and explainability (3-5), to evaluating whether an AI system meets the design criteria of accountability and respect for privacy (6-8), to analyzing whether an AI system satisfies the full set of ethical design criteria: fairness, transparency, explainability, accountability, respect for privacy, and adherence to societal values (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)Recognise that AI applications directly interacting with humans and making decisions about their lives (e.g. CV-sorting, exam scoring) can be controversial even where other AI applications are not (AI example 63, competence 2.3).
- UNESCOEvaluate the consequences of institutional use of AI to make high-stakes decisions about people (e.g. profiling students, hiring, loan/assessment scoring) and understand why human accountability must not be ceded to AI in such decisions (CG4.2.1.2).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- ChinaStudents build a three-dimensional 'technical principle → system architecture → social impact' way of reasoning about AI
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
Deepfakes and Misinformation25 lessons· meets 39 requirements inUAEADEKIBEngland + Cambridge+7
Lessons
- The Impossible Video
- What Are Deepfakes?
- The Viral Clip Puzzle
- Types of Synthetic Media
- Why Fakes Work So Fast
- Which Generator?
- Match the Fake Types!
- Why Detection Keeps Chasing Generation
- Dangers of Deepfakes
- Sort the Harm Types!
- How to Spot Deepfakes
- Clue or Proof?
- Combating Deepfakes
- Estimate Deepfake Risk
- Why Platforms Label Synthetic Media
- Tool for the Job
- Responding to Fakes
- Same Facts, Different Frame
- Neutral or Steering?
- Framing Check
- Spot the Red Flags!
- Deepfake Expert Quiz
- Explain Why Verification Comes First
- Deepfake Vocabulary
- Truth Defender!
Requirements this topic helps meet
- UAEImprove AI systems to match societal expectations
- UAEAI and human expression, behaviour, creativity
- UAEApply effective prompting strategies
- ADEKGAN/diffusion image generation, critique an image
- ADEKDeepfakes: explain, spot flaws, set responsible-use rules
- ADEKDraft/commit to ethical-use guidelines
- IBUnderstand the bias inherent in content produced by an AI tool and be able to critically review AI-generated output
- England + CambridgeUnderstand how changes in technology affect safety, including new ways to protect online privacy and identity, and how to identify and report a range of concerns
- England + CambridgeMedia literacy (taught mainly via Citizenship) should cover critically engaging with messages conveyed through different media channels, including AI, and identifying/protecting against misinformation and disinformation
- CSTA (US)Evaluate and refine computational artifacts to make them more usable and accessible.
- CSTA (US)Evaluate the ways computing impacts personal, ethical, social, economic, and cultural practices.
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how AI shapes beliefs and behaviours (Engage 1, advanced level).
- OECD–EC (Europe)Evaluate whether an AI output should be accepted, revised, or rejected, and justify the decision (Engage 3, advanced).
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Weigh the benefits and disadvantages of using AI-driven search engines, including privacy and commercial-interest trade-offs (AI example 31, competence 1.1).
- OECD–EC (Europe)Be aware that AI algorithms may embody a commercial or political message rather than only serving the user's information need, with possible negative consequences such as reproducing stereotypes or spreading misinformation (AI example 48, competence 1.2).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)Know that 'deep-fakes' are AI-generated images, video, or audio of events or people that did not really happen, and may be indistinguishable from the real thing (AI example 51, competence 1.2).
- UNESCOCarry out responsible AI practices in compliance with ethical principles and locally applicable regulations, iteratively building and following a personal checklist of lawful and responsible practices (e.g. protecting privacy/copyright, marking AI-generated content, avoiding disinformation, hate speech, or sensitive content about identifiable individuals) (CG4.2.2.1).
- UNESCOGain an age-appropriate understanding of personal, legal and ethical responsibilities when using AI and reinforce self-disciplined behaviours around sensitive personal data, copyrighted materials, deepfake/identifiable-person imagery, and the spread of misinformation, disinformation or hate speech (CG4.2.2.2).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- SingaporeDevelop digital safety and security competencies as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital responsibility as one of the nine core NDLP digital competencies.
- SingaporeLearn to use AI: use AI tools safely and responsibly.
- SingaporeThink critically about AI's biases, question AI-generated outputs, and discern or identify false or misleading information produced by AI.
- Korea, Japan, HKUnderstand laws/regulations on information, the importance of information security, and individual responsibility and information morals in the information society.
- Korea, Japan, HKScientifically understand the characteristics of media and communication methods, including how they have changed over time, and the role information design plays for people and society.
- Korea, Japan, HKUnderstand and apply the concepts and methods of information design (including universal design, usability and accessibility) needed for effective communication with a given audience, and evaluate/improve created content.
- Korea, Japan, HKFindings from FY2023 pilot schools showed some students uncritically accepted AI answers as fact or could not judge appropriate situations for use on their own — confirming an explicit target competency: judge the appropriateness/veracity of generative-AI output rather than accept it uncritically.
- Australia + NZDesign the user experience of a digital system generating or processing data, and evaluate the suitability of user interfaces for defined purposes and end-users, applying accessibility standards for diverse audiences
- Australia + NZStudents develop an understanding of how generative AI tools work in broad terms, how they can be used appropriately, and when/how such tools are affecting the content and information they see (transparency about AI-generated or AI-assisted content)
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
AI and the Future of Work22 lessons· meets 41 requirements inADEKCBSECISCE + statesIB+8
Lessons
- The Robot Takeover?
- Jobs AI Changes
- The Job Change Puzzle
- What AI Automates: Tasks, Not Jobs
- Jobs Are Bundles of Tasks
- Audit a Job
- Job Impact Match
- Why Hybrid Roles Keep Growing
- Automate, Assist or Lead?
- Jobs AI Creates
- Growing vs Shrinking
- Skills of Tomorrow
- The Big Picture
- Model a Future Job
- Why Designers Use AI but Still Lead
- Read the Forecast
- Job Transformation
- Spot AI at Work
- Jobs Quiz
- Explain Why AI Changes More Jobs Than It Replaces
- Work Words
- Future Ready!
Requirements this topic helps meet
- ADEKDraft/commit to ethical-use guidelines
- ADEKCareers, future of work
- CBSEComplete Unit 3: Information and Communication Technology Skills (Level I-IV progressing Class IX to XII), covering topics such as email introduction and account creation, and networking basics at Class IX per secondary sources.
- CISCE + statesExplain what a robot is, its characteristics, and the benefits of robots over humans (quality, productivity, efficiency, working span, hazardous environments)
- CISCE + statesState a brief history of robotics evolution since 1900 and recite Isaac Asimov's Three Laws of Robotics
- CISCE + statesClassify robots by terrain/field (aerial, ground, underwater), by control (manual, automatic), and identify bio-inspired robots (humanoid, bird, snake, insect)
- CISCE + statesGive at least one real-world application example of robots in domestic, industry, medical, defence, entertainment and agriculture domains
- CISCE + statesExplain New Age Robotic Systems (warehouse, assistant, smart-home/school/mobility robots, autonomous cars/drones, medical/healthcare robotics) and why NARS are becoming increasingly relevant and possible
- CISCE + statesDifferentiate a machine from a robot, and a robot from a cobot (human-in-the-loop, multi-robot systems), tracing the progress from robots to cobots
- IBUse and apply technology effectively to access, process, and communicate information, and to model and create solutions to problems
- IBDevelop an appreciation of the impact of design and technology innovations on life, global society, and environments, including their historical, political, and social contexts
- England + CambridgeDevelop capability, creativity and knowledge in computer science, digital media and information technology (statutory for all KS4 pupils, not only GCSE candidates)
- England + CambridgeUnderstand what is meant by robotics, describe the characteristics of a robot (mechanical structure, electrical components/sensors/actuators/microprocessors, programmability), and the roles robots perform (industry, transport, agriculture, medicine, domestic, entertainment) with advantages/disadvantages
- CSTA (US)Evaluate the ways computing impacts personal, ethical, social, economic, and cultural practices.
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from identifying current uses of AI and how they have impacted society (K-2), to identifying how sectors of society such as manufacturing, retail, and healthcare are changing due to AI (3-5), to comparing AI-driven societal change to previous industrial revolutions (6-8), to predicting how a specific sector of society is likely to change in the short and intermediate term as a result of AI, referencing Amara's Law (9-12).
- AI4K12 (US)Progress from describing jobs that no longer exist due to past technology advances, e.g. farriers replaced by auto mechanics (K-2), to describing how a specific job will change due to AI/robotic technology, e.g. warehouse workers alongside robots (3-5), to predicting a new job type that might arise or an existing job that might change/disappear due to AI adoption (6-8), to investigating the specific skills needed for AI-enabled careers across the workforce, not just for programmers (9-12).
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Deliberately assign parts of a task to human strengths versus AI capabilities (Manage 3, advanced).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)Be aware that AI-based technologies can replace some human functions, potentially causing job losses or reallocations while also creating new jobs (AI example 60, competence 2.4).
- UNESCOExplain that AI is human-led and that decisions by AI creators shape how AI systems affect human rights, human-AI interaction, and society; recognise that AI cannot replace human thinking or intellectual development (CG4.1.1.1).
- UNESCOCritically examine the dynamic relationship between human agency and machine agency, including cases where a degree of machine agency may be warranted (e.g. medical pattern detection) versus cases requiring human control (CG4.1.1.3).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- UNESCOReflect on personal goals and a lifelong-learning, self-actualization attitude toward AI as it becomes pervasive in work, life and social practices (CG4.3.1.3).
- SingaporeKeep digital knowledge current, including awareness of emerging technologies such as AI, as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital information management skills as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital communication, collaboration and participation skills as one of the nine core NDLP digital competencies.
- SingaporeApply the Find-Think-Apply-Create framework when locating, evaluating, using and creating with digital information (exact breakdown of the four components not published on the primary MOE page reviewed).
- SingaporeLearn beyond AI: develop distinctly human capabilities and dispositions that AI cannot replicate.
- SingaporeDevelop digital literacy and technological skills needed to thrive in an increasingly technology-driven world, as a central goal of the Masterplan.
- Korea, Japan, HKUnderstand the properties of information and media, and use information and information technology to discover and solve problems in the information society.
- Korea, Japan, HKUnderstand the roles information technology plays for people and society and its effects, explicitly including how AI and robots change the nature of work and raise cybercrime/digital-divide/health concerns.
- Australia + NZApply digital literacy skills (investigating with, creating with, communicating and collaborating with, and managing/operating digital tools) across all learning areas, including recognising the role of AI-enabled tools in everyday digital systems
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
AI Governance and Regulation22 lessons· meets 25 requirements inUAEADEKCSTA (US)AI4K12 (US)+4
Lessons
- Who Controls AI?
- The Rule Makers
- Who Writes Which Rule?
- Same AI, Different Rules
- Why AI Needs Rules
- One Tool, Many Rulebooks
- Principles Match
- Why Risk Changes Everything
- AI Laws Around World
- AI Risk Levels
- Ethics Boards
- Real Board or Ethics Washing?
- Governance Challenges
- Governance Under Pressure
- Test a Governance Plan
- Why ChatGPT Needs Policy Rules
- Making AI Rules
- Governance Check
- Governance Quiz
- Explain Risk-Based Governance
- Governance Terms
- Rule Maker!
Requirements this topic helps meet
- UAEEffectiveness of AI governance
- ADEKFacial recognition: privacy/safety/fairness tension
- ADEKDraft/commit to ethical-use guidelines
- ADEKAI governance/civic participation
- CSTA (US)Evaluate the social and economic implications of privacy in the context of safety, law, or ethics.
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Establish checkpoints, monitor AI use against criteria, and adjust roles throughout a problem-solving process (Manage 4, advanced).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)Know that processing of personal data by AI (e.g. voice interactions with a virtual assistant) is subject to regulations such as the EU GDPR and can carry data protection, privacy and security risks (AI example 39, competence 4.2).
- UNESCOUnderstand the necessity of exercising sufficient human control over AI, including the consequences when human oversight fails or is absent (CG4.1.1.2).
- UNESCOUnderstand that human accountability is a legal obligation of AI creators and AI service providers, who are answerable for safety incidents, ethical risks in design/training, and misuses of their AI service (CG4.2.1.1).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- UNESCOCultivate social responsibilities to uphold 'ethics by design' in regulations on AI — evaluating how selected local AI regulations align with ethics-by-design, and drafting proposals to modify or create new regulations governing ethics by design (CG4.3.2.3).
- UNESCOBuild technical and social-responsibility skills to make decisions on optimizing, reconfiguring, or shutting down an AI system based on test results and user feedback (three scenarios: optimize datasets/model/interface; reconfigure problem-scoping/architecture; or shut down and remediate where the system violates human rights or harms vulnerable groups) (CG4.3.4.2).
- UNESCONurture a self-identity as a co-creator and 'driver' of the next generation of AI technology, critically analyse the long-term impacts of AI systems on social relations and behaviours, and discuss how regulations/policies should adapt to enhance AI governance (e.g. by engaging with local/online communities of AI co-creators) (CG4.3.4.3).
- Korea, Japan, HKStudents and teachers should treat generative AI output as one reference among others, always keeping human judgement and responsibility final, and use the information only after checking it against the qualities/abilities set out in the curriculum guidelines.
- Australia + NZEvaluate how the Australian Privacy Principles apply to the storage and communication of data in digital systems, and assess/manage the risks of common cyberthreats (e.g. phishing, malware) to a system
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
AI for Global Good22 lessons· meets 28 requirements inADEKCBSECISCE + statesEngland + Cambridge+7
Lessons
- AI Goes Global
- AI for All
- One Phone, Many Kinds of Help
- Global Health AI
- One Model, Many Communities
- Lab vs Clinic
- Global AI Match
- Why Access Beats Hype
- AI for Farmers
- AI Focus Areas
- AI Education
- Global AI Challenges
- Barrier to Fix
- Design for Global Reach
- How Google Translate Scales Help
- Fair for Whom?
- AI for Good Steps
- AI World Tour
- Global AI Quiz
- Explain Inclusive AI Design
- Global AI Words
- World Changer!
Requirements this topic helps meet
- ADEKDraft/commit to ethical-use guidelines
- CBSEComplete a Part D project/field visit/student portfolio related to Sustainable Development Goals, e.g. building an AI model with Teachable Machine or Machine Learning for Kids, or maintaining a portfolio of AI activities (4Ws canvas, system map, research on AI for SDGs).
- CBSEEmpower students to create positive change through AI-driven social impact projects, and equip students with skills to develop AI solutions addressing societal challenges (course objectives).
- CISCE + statesGive at least one real-world application example of robots in domestic, industry, medical, defence, entertainment and agriculture domains
- CISCE + statesIdentify applications and benefits of AI across commercial, industry, medical/healthcare, defence, banking, entertainment, transport, security and agriculture domains
- CISCE + statesExplain New Age Robotic Systems (warehouse, assistant, smart-home/school/mobility robots, autonomous cars/drones, medical/healthcare robotics) and why NARS are becoming increasingly relevant and possible
- England + CambridgeYoung people should understand how AI works, its capabilities and limitations, and learn to use it effectively without becoming dependent on it; this should be delivered primarily through Computing rather than assumed to develop automatically
- CSTA (US)Evaluate and refine computational artifacts to make them more usable and accessible.
- CSTA (US)Evaluate the ways computing impacts personal, ethical, social, economic, and cultural practices.
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from describing how AI can be used to solve a societal problem, e.g. Code.org's AI for Oceans classifying wildlife vs. debris (K-2), to designing a solution to a societal problem using AI technology, e.g. environmental protection, energy conservation (3-5), to researching a societal problem and describing how AI perception/reasoning/learning technologies could address it, referencing the UN Sustainable Development Goals (6-8), to evaluating an actual AI-for-social-good project in terms of the problem addressed and its real or potential impact (9-12).
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Assess an AI system's strengths and limitations considering its purpose and potential impacts (Shape 1, advanced).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)States that a foundational understanding of AI's technical functioning and the ability to recognise chances, limits and risks are indispensable for students to use AI responsibly and reflectively ('digital mündig'); calls on the Länder to integrate AI-related informatics competences and 'Bildung in der digitalen Welt' competences into curricula for all subjects and school types, without itself specifying discrete grade-level student outcomes.
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- SingaporeUnderstand AI's uses, risks and limitations, and apply ethical considerations (e.g. academic integrity, proper data handling) when using AI to support learning.
- SingaporeIdentify and explain AI's potential benefits, limitations and risks.
- SingaporeGive examples of common personal and business tasks that can be performed well by AI: face recognition, voice recognition, image classification and spam filtering.
- Korea, Japan, HKUnderstand and apply the concepts and methods of information design (including universal design, usability and accessibility) needed for effective communication with a given audience, and evaluate/improve created content.
- Australia + NZDesign the user experience of a digital system generating or processing data, and evaluate the suitability of user interfaces for defined purposes and end-users, applying accessibility standards for diverse audiences
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
Responsible AI Development36 lessons· meets 39 requirements inADEKIBCSTA (US)AI4K12 (US)+5
Lessons
- Building AI Right
- Core Principles
- Why Good Intentions Are Not Enough
- Data Responsibility
- Documentation Changes Behavior
- Practice Match
- Why Testing Beats Confidence
- Testing for Fairness
- Read the Split Results
- Build a Biased AI on Purpose
- Run It on Your AI
- What to Look For
- Your Own Bias Investigation
- Read Your Experiment
- My Before and After
- Good vs Bad Practice
- Model Cards
- Fill the Model Card
- Diverse Teams Win
- Tune a Responsible Build
- Why Copilots Need Guardrails
- Which Risk Tier?
- Development Steps
- Responsibility Check
- Capstone: Your Companion's Model Card
- Train, Measure, Read the Card
- Read the Metrics
- Secure or Not Yet?
- Defend Your Weakest Number
- Two More Lines: Explainability and Privacy
- Explainability or Privacy?
- Write Both Lines
- Development Quiz
- Explain Responsible Development
- Dev Terms
- Responsible Builder!
Requirements this topic helps meet
- ADEKDraft/commit to ethical-use guidelines
- ADEKAI governance/civic participation
- IBPlan the creation of the chosen solution and follow that plan to build a prototype sufficient for testing and evaluation
- CSTA (US)Test and refine computational artifacts to reduce bias and equity deficits.
- CSTA (US)Evaluate the social and economic implications of privacy in the context of safety, law, or ethics.
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate and analyze the roles of humans and human decision-making in the creation of AI.
- CSTA (US)Evaluate an AI model (e.g., using a model card) to determine the model's features as well as its biases, explainability, fairness, privacy, accuracy, and transparency.
- CSTA (US)Train, iteratively improve, and then develop a model card for an AI model with ethical considerations for end users and others impacted: fairness, bias, safety, security, intellectual property, privacy, robustness, explainability, accuracy, transparency, and accountability.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- CSTA (US)Students read, evaluate, and critique AI-generated code as part of the Programming strand.
- AI4K12 (US)Progress from giving examples of intelligent vs. non-intelligent machines and discussing what makes a machine 'intelligent' (K-2), to demonstrating a machine perception tool (e.g. speech transcription, object recognition) and explaining why it is perception rather than mere sensing (3-5), to giving examples of specialized perception algorithms (face detection, obstacle detection, speech recognition, etc.) (6-8), to explaining how perception algorithms power real-world applications like license plate readers and Google Lens (9-12).
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from creating a model card for a classification or prediction model documenting its intended use, training data, accuracy, and limitations (3-5), to constructing an ethical matrix of stakeholders and their values for an AI system and evaluating alignment (6-8), to designing an AI system using a full ethical design process that considers all stakeholders' values across fairness, transparency, explainability, accountability, privacy, and societal values (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from describing and using AI extensions/plugins in a familiar programming framework such as Scratch, e.g. speech-to-text, face recognition (3-5), to creating a novel application using AI extensions/plugins in a chosen framework such as Scratch, MIT App Inventor, or Calypso (6-8), to creating a novel application using AI tools of the student's choice, now commonplace and freely available without advanced degrees or expensive equipment (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Establish checkpoints, monitor AI use against criteria, and adjust roles throughout a problem-solving process (Manage 4, advanced).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)Know that processing of personal data by AI (e.g. voice interactions with a virtual assistant) is subject to regulations such as the EU GDPR and can carry data protection, privacy and security risks (AI example 39, competence 4.2).
- UNESCOUnderstand the necessity of exercising sufficient human control over AI, including the consequences when human oversight fails or is absent (CG4.1.1.2).
- UNESCOUnderstand human roles across the AI life cycle (researchers, architecture/data engineers, beta testers, ethics/safety regulators, human-AI interface specialists, compliance auditors) and the main ethical issues in using data to train AI systems (CG4.1.3.4).
- UNESCOCategorise general safety risks of AI and typical AI incidents, deepen knowledge of human rights to data protection and privacy and creators' legal responsibility to collect data only with informed consent, and practise strategies to ensure personal data is collected, used, shared, archived and deleted only with consent (CG4.2.2.3).
- UNESCOCritically evaluate and leverage free/open-source AI datasets (e.g. MNIST, CIFAR, ImageNet) and AI/algorithm libraries (e.g. Teachable Machine, PyTorch, Keras) to address authentic, task-based problems (CG4.2.3.3).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- UNESCODevelop a critical attitude to ethics-by-design principles behind existing AI systems/algorithms by auditing selected tools, proposing an ethics label, and recommending remedies for identified violations of ethical principles or regulations (CG4.3.2.2).
- UNESCOTransfer values, knowledge and skills to craft a task-based AI tool by customizing an existing AI model or toolkit, including modifying/creating a dataset for real-world contexts (e.g. scraping data with tools like BeautifulSoup, cleaning/encoding/preprocessing it) (CG4.3.3.1).
- UNESCOEnhance creativity in customizing AI toolkits and coding — leveraging AI development platforms, enhancing datasets, and modifying programming code (including open-source options) to design and test creative task-based AI tools (CG4.3.3.2).
- UNESCOPractise critiquing AI systems by auditing whether a model enhances or weakens human capacities, agency and consciousness, checking the extent of its explainability and data-privacy protection, and measuring system performance and its broader societal/environmental impact (CG4.3.4.1).
- UNESCOBuild technical and social-responsibility skills to make decisions on optimizing, reconfiguring, or shutting down an AI system based on test results and user feedback (three scenarios: optimize datasets/model/interface; reconfigure problem-scoping/architecture; or shut down and remediate where the system violates human rights or harms vulnerable groups) (CG4.3.4.2).
- Korea, Japan, HKStudents and teachers should treat generative AI output as one reference among others, always keeping human judgement and responsibility final, and use the information only after checking it against the qualities/abilities set out in the curriculum guidelines.
- ChinaStudents build a three-dimensional 'technical principle → system architecture → social impact' way of reasoning about AI
- Australia + NZEvaluate how the Australian Privacy Principles apply to the storage and communication of data in digital systems, and assess/manage the risks of common cyberthreats (e.g. phishing, malware) to a system
- Australia + NZRecognise how the identification and mitigation of bias is a critical aspect of machine learning, and critically evaluate the reliability, validity and ethics of AI-generated outputs and predictive strategies
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
Building Your AI Ethics Framework24 lessons· meets 43 requirements inUAEADEKCBSECISCE + states+8
Lessons
- Your AI Compass
- Your Core Values
- The Same Tool, Different Choice
- Questions to Ask
- A Framework Travels With You
- Checklist on a Case
- Values in Action
- Why a Pause Creates Clarity
- Reason or Reflex?
- Decision Process
- Ethical Responses
- Your AI Roles
- Actions You Can Take
- Tune Your Ethics Compass
- Running TikTok Through a Framework
- EU AI Act Risk Tiers
- Framework Steps
- Framework in Action
- Framework Quiz
- Explain Your Ethics Framework
- Framework Terms
- Judge the Hiring Case
- Write the Rule
- Ethics Champion!
Requirements this topic helps meet
- UAEJudge responsible AI use against governance frameworks
- ADEKDeepfakes: explain, spot flaws, set responsible-use rules
- ADEKStakeholder debate / ethical matrix
- ADEKDraft/commit to ethical-use guidelines
- ADEKAI governance/civic participation
- CBSEUnderstand the stages of the AI Project Cycle and the three AI domains with illustrations of practical applications; learn about ethical frameworks for AI and their categories, including Bioethics as a framework used in healthcare (Unit 1).
- CISCE + statesDescribe ethical considerations in AI including bias, prejudice, fairness, accountability, transparency, interpretability and explainability
- IBAct with integrity and honesty and take responsibility for one's own actions when making design decisions
- IBUnderstand that any AI-generated work included, even in part, in a piece of assessment is not considered the student's own unless it is clearly credited in the text and referenced in the bibliography
- IBFollow the IB Academic Integrity policy's guidance on the ethical use of AI tools (per its dedicated appendix) when producing assessed work
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate the design, motivation, outcomes, and potential impacts of AI systems using ethical design criteria and/or ethical frameworks.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from discussing what makes a system fair vs. unfair through a fair-distribution exercise (K-2), to evaluating whether an AI system meets the design criteria of transparency and explainability (3-5), to evaluating whether an AI system meets the design criteria of accountability and respect for privacy (6-8), to analyzing whether an AI system satisfies the full set of ethical design criteria: fairness, transparency, explainability, accountability, respect for privacy, and adherence to societal values (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how AI shapes beliefs and behaviours (Engage 1, advanced level).
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Evaluate how well an AI use aligns with ethical principles and human values across trade-offs (Engage 7, advanced).
- OECD–EC (Europe)Assess the ethical implications and justify how an AI-assisted work maintains authenticity (Create 4, advanced).
- OECD–EC (Europe)Weigh the benefits and disadvantages of using AI-driven search engines, including privacy and commercial-interest trade-offs (AI example 31, competence 1.1).
- OECD–EC (Europe)Be aware that AI algorithms may embody a commercial or political message rather than only serving the user's information need, with possible negative consequences such as reproducing stereotypes or spreading misinformation (AI example 48, competence 1.2).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)Know that 'deep-fakes' are AI-generated images, video, or audio of events or people that did not really happen, and may be indistinguishable from the real thing (AI example 51, competence 1.2).
- UNESCOIllustrate dilemmas around AI and identify the main reasons behind ethical conflicts that AI creators and companies face (e.g. data collection vs. ownership, profit vs. privacy, safety vs. speed of iteration) (CG4.1.2.1).
- UNESCODiscuss age-appropriate real-world cases around six core AI ethical principles — do no harm, proportionality, non-discrimination, sustainability, human determination, and transparency/explainability — and their personal implications (CG4.1.2.2).
- UNESCOInternalise ethical principles on AI (human rights, data privacy, human agency, equity, inclusion, social justice, environmental sustainability) into reflective practices, e.g. advocating for inclusive AI interfaces and reporting discriminatory bias found in AI tools (CG4.1.2.3).
- UNESCOCarry out responsible AI practices in compliance with ethical principles and locally applicable regulations, iteratively building and following a personal checklist of lawful and responsible practices (e.g. protecting privacy/copyright, marking AI-generated content, avoiding disinformation, hate speech, or sensitive content about identifiable individuals) (CG4.2.2.1).
- UNESCOGain an age-appropriate understanding of personal, legal and ethical responsibilities when using AI and reinforce self-disciplined behaviours around sensitive personal data, copyrighted materials, deepfake/identifiable-person imagery, and the spread of misinformation, disinformation or hate speech (CG4.2.2.2).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- UNESCOArticulate personal and social responsibilities of citizens in an AI society — from both user and designer perspectives — including ensuring human control and accountability over every step of the AI life cycle (CG4.3.1.2).
- UNESCOCultivate social responsibilities to uphold 'ethics by design' in regulations on AI — evaluating how selected local AI regulations align with ethics-by-design, and drafting proposals to modify or create new regulations governing ethics by design (CG4.3.2.3).
- UNESCONurture a self-identity as a co-creator and 'driver' of the next generation of AI technology, critically analyse the long-term impacts of AI systems on social relations and behaviours, and discuss how regulations/policies should adapt to enhance AI governance (e.g. by engaging with local/online communities of AI co-creators) (CG4.3.4.3).
- SingaporeLearn to use AI: use AI tools safely and responsibly.
- ChinaStudents practice social responsibility, weighing technological innovation against societal risk when using or designing AI
- Australia + NZStudents understand that generative AI tools should be used in ways that benefit all members of the school community and support wellbeing, not just individual convenience
- Australia + NZStudents develop an understanding of how generative AI tools work in broad terms, how they can be used appropriately, and when/how such tools are affecting the content and information they see (transparency about AI-generated or AI-assisted content)
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
- Australia + NZStudents learn that use of generative AI outputs (e.g. in assessable work) must remain open to challenge, retain human agency, and that the student/user remains accountable for decisions made with AI assistance (e.g. academic integrity implications)
- Australia + NZStudents understand that their privacy and personal data must be protected when using generative AI tools, and are taught not to input sensitive personal or others' data into such tools without appropriate safeguards
AI Ethics Unit Review22 lessons· meets 29 requirements inADEKCBSEIBEngland + Cambridge+8
Lessons
- Ethics Journey
- Unit Topics
- One Lens for Many AI Problems
- Big Ideas
- Case File: The Library Scanner
- One Unit, One Bigger Pattern
- Unit Mega Match
- Why Review Matters
- Your Tools
- Run the Numbers Again
- Ethics Check
- Future of AI Ethics
- Forever Lessons
- Which Principle Is Under Pressure?
- Balance an Ethics Review
- Why YouTube Needs Multiple Safeguards
- PAUSE in Practice
- Final Challenge
- Unit Final Quiz
- Explain the Core Ethics Pattern
- Unit Vocabulary
- AI Ethics Graduate!
Requirements this topic helps meet
- ADEKDraft/commit to ethical-use guidelines
- CBSEComplete Unit 3: Information and Communication Technology Skills (Level I-IV progressing Class IX to XII), covering topics such as email introduction and account creation, and networking basics at Class IX per secondary sources.
- IBUse and apply technology effectively to access, process, and communicate information, and to model and create solutions to problems
- IBAct with integrity and honesty and take responsibility for one's own actions when making design decisions
- England + CambridgeDevelop capability, creativity and knowledge in computer science, digital media and information technology (statutory for all KS4 pupils, not only GCSE candidates)
- CSTA (US)Evaluate the social and economic implications of privacy in the context of safety, law, or ethics.
- CSTA (US)Evaluate computational artifacts to maximize their beneficial effects and minimize harmful effects on society.
- CSTA (US)Evaluate the impact of equity, access, and influence on the distribution of computing resources in a global society.
- CSTA (US)Debate laws and regulations that impact the development and use of software.
- CSTA (US)Evaluate how AI use impacts an individual's decision making and other behavior.
- CSTA (US)Evaluate the intended and unintended impacts of AI on society (e.g., deep fakes, job loss), including government, education, entertainment, culture, careers, and national security, while considering how these impacts may differ among diverse communities.
- AI4K12 (US)Progress from evaluating the ways a decision impacts different people differently (K-2), to evaluating how an AI technology, e.g. speech or facial recognition, has different effects on different demographic groups (3-5), to evaluating how various stakeholders' conflicting goals and values shape AI system design decisions (6-8), to explaining how the use of AI systems (e.g. health-risk algorithms) has led to documented disparate impacts on different groups (9-12).
- AI4K12 (US)Progress from analyzing deepfake images/videos to identify the flaws that currently reveal them, e.g. thispersondoesnotexist.com (3-5), to critiquing uses of AI technology for surveillance or privacy violation, e.g. face recognition, license-plate trackers (6-8), to identifying appropriate areas for AI regulation and evaluating proposed legislation, e.g. autonomous-vehicle laws (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how institutional deployment of AI can serve particular interests and amplify societal bias (Engage 6, advanced).
- OECD–EC (Europe)Be aware that the data on which AI depends may include biases which can be automated and worsened by AI use (AI example 49, competence 1.2).
- OECD–EC (Europe)Know that processing of personal data by AI (e.g. voice interactions with a virtual assistant) is subject to regulations such as the EU GDPR and can carry data protection, privacy and security risks (AI example 39, competence 4.2).
- UNESCOGain evidence-based insight into the pervasive adoption of AI as infrastructure of social activity, and critique challenges AI poses to inclusion, justice and sustainability, including AI-amplified bias against women, marginalised ethnic groups and socio-economically disadvantaged people (CG4.3.1.1).
- UNESCOArticulate personal and social responsibilities of citizens in an AI society — from both user and designer perspectives — including ensuring human control and accountability over every step of the AI life cycle (CG4.3.1.2).
- SingaporeKeep digital knowledge current, including awareness of emerging technologies such as AI, as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital information management skills as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital communication, collaboration and participation skills as one of the nine core NDLP digital competencies.
- SingaporeApply the Find-Think-Apply-Create framework when locating, evaluating, using and creating with digital information (exact breakdown of the four components not published on the primary MOE page reviewed).
- SingaporeDevelop digital literacy and technological skills needed to thrive in an increasingly technology-driven world, as a central goal of the Masterplan.
- Korea, Japan, HKUnderstand the properties of information and media, and use information and information technology to discover and solve problems in the information society.
- ChinaStudents practice social responsibility, weighing technological innovation against societal risk when using or designing AI
- Australia + NZEvaluate how the Australian Privacy Principles apply to the storage and communication of data in digital systems, and assess/manage the risks of common cyberthreats (e.g. phishing, malware) to a system
- Australia + NZApply digital literacy skills (investigating with, creating with, communicating and collaborating with, and managing/operating digital tools) across all learning areas, including recognising the role of AI-enabled tools in everyday digital systems
- Australia + NZStudents understand that generative AI tools should be used in ways that benefit all members of the school community and support wellbeing, not just individual convenience
- Australia + NZStudents are taught that generative AI tools must be used, and their outputs treated, in ways that are accessible, fair and respectful to all users, being alert to potential bias in AI outputs
Keeping Secrets: Encryption20 lessons· meets 15 requirements inIBEngland + CambridgeCSTA (US)Singapore+2
Lessons
- The Café Wi-Fi Question
- Plaintext, Key, Ciphertext
- Run the Shift Cipher
- Crack Byte's Messages
- XOR on a Byte
- XOR by Hand
- The Key-Sharing Problem
- Public-Key Encryption with Tiny RSA
- Encrypt with the Public Key
- Build an RSA Key Pair
- Hashing Is Not Encryption
- Encrypt or Hash?
- Inside the HTTPS Padlock
- Why AI Needs Encryption
- Protected or Exposed?
- Hash Chains and Blockchains
- Guaranteed or Not?
- Checkpoint: Keeping Secrets
- Encryption Vocabulary
- Secret Keeper!
Requirements this topic helps meet
- IBDescribe network structures, protocols, and security threats (encryption, firewalls, phishing, DDoS attacks)
- England + CambridgeUnderstand how changes in technology affect safety, including new ways to protect online privacy and identity, and how to identify and report a range of concerns
- England + CambridgeUnderstand the difference between the internet and the world wide web, and describe how web pages are located, retrieved and displayed (browser, IP addresses, DNS, web server, HTML)
- England + CambridgeUnderstand the concept of a digital currency and the process of blockchain as a time-stamped digital ledger used to track digital currency transactions
- England + CambridgeShow understanding of encryption (symmetric/asymmetric cryptography, public/private key), SSL/TLS, digital certification and quantum cryptography purpose/benefits/drawbacks
- CSTA (US)Apply multiple methods of encryption to model the secure transmission of information.
- CSTA (US)Evaluate the scalability and reliability of networks, by describing the relationship between routers, switches, servers, topology, and addressing.
- CSTA (US)Describe the issues that impact network functionality (e.g., bandwidth, load, delay, topology).
- SingaporeDevelop digital safety and security competencies as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital responsibility as one of the nine core NDLP digital competencies.
- Korea, Japan, HKUnderstand laws/regulations on information, the importance of information security, and individual responsibility and information morals in the information society.
- Korea, Japan, HKUnderstand the structure and components of communication networks, the role of protocols, and methods/techniques for ensuring information security; design a small-scale network.
- Australia + NZExamine network protocols, and how networks apply cybersecurity measures to protect data, software and hardware
- Australia + NZExplain how data compression and encryption techniques are used to store and transmit data securely and efficiently
- Australia + NZStudents understand that their privacy and personal data must be protected when using generative AI tools, and are taught not to input sensitive personal or others' data into such tools without appropriate safeguards
Attacks on AI Systems25 lessons· meets 10 requirements inUAEEngland + CambridgeCSTA (US)Australia + NZ
Lessons
- The Homework Bot
- Thinking Like an Attacker
- Phishing, Now Written by AI
- Phishing or Legit?
- Prompt Injection
- Direct, Indirect or Neither?
- Poisoning the Well
- Move the Line
- Stealing a Model
- Adversarial Examples
- Match the Attack
- Spot the Attack
- Proportionate Defences
- Attack to Defence
- Does It Fit the Risk?
- Build a Threat Model
- The Chatbot That Learned Too Well
- Defending With AI
- Which Defence Spots It?
- The Attacker-Defender Race
- Checkpoint: Defending With AI
- Checkpoint: Attacks and Defences
- Explain Proportionate Defence
- Security Terms
- Security Minded!
Requirements this topic helps meet
- UAEHuman–AI collaboration across sectors
- England + CambridgeDescribe the processes involved in, and aim of, a range of cyber security threats (brute-force, data interception, DDoS, hacking, malware, pharming, phishing, social engineering) and explain solutions to keep data safe (access levels, anti-malware, authentication, firewalls, SSL, etc.)
- CSTA (US)Explain security issues that might lead to compromised computer programs.
- CSTA (US)Explain how physical and digital security measures protect electronic information.
- CSTA (US)Give examples to illustrate how sensitive data can be affected by malware and other attacks.
- CSTA (US)Recommend security measures to address various scenarios based on factors such as efficiency, feasibility, and ethical impacts.
- CSTA (US)Compare various security measures, considering tradeoffs between the usability and security of a computing system.
- CSTA (US)Explain tradeoffs when selecting and implementing cybersecurity recommendations.
- CSTA (US)Compare ways software developers protect devices and information from unauthorized access.
- Australia + NZStudents understand that generative AI tools should be used in ways that benefit all members of the school community and support wellbeing, not just individual convenience
Choosing and Judging AI Tools30 lessons· meets 36 requirements inUAEADEKCBSEIB+7
Lessons
- Which AI Is Best?
- Best at What?
- Which Question Matters Most?
- Same Question, Different Answers
- A Fair Reliability Test
- Averages Can Hide Failures
- Try It: Run a Reliability Test
- What Did Your Test Show?
- Fair Test or Not?
- Models Built Around the World
- Who Built It?
- A Model That Reads Arabic
- Side by Side: Four Models, One Task
- Score the Four
- Choose a Tool
- Try It: Compare Two Real Models
- Your Work, Your Thinking
- Coach or Ghostwriter?
- Say How You Used AI
- Enough Detail?
- Plan With AI, Then Check the Plan
- Keep or Fix?
- Put the Plan in Order
- Your Turn: Plan and Message With AI
- My Plan and Message
- Six Cases That Didn't Exist
- Checkpoint: Choosing and Judging AI Tools
- Explain Your Choice
- Tool-Judging Terms
- Tool Tester!
Requirements this topic helps meet
- UAEAI and human expression, behaviour, creativity
- UAEUse AI tools to work more effectively
- UAEApply effective prompting strategies
- ADEKAI governance/civic participation
- CBSEComplete Unit 3: Information and Communication Technology Skills (Level I-IV progressing Class IX to XII), covering topics such as email introduction and account creation, and networking basics at Class IX per secondary sources.
- IBCredit and reference any AI-generated content used in coursework in the body of the text and in the bibliography, per IB academic integrity requirements
- IBUse and apply technology effectively to access, process, and communicate information, and to model and create solutions to problems
- IBWhere a school offers digital design (rather than or alongside product design), demonstrate digital-tool competencies such as prototyping, graphic/web design, and multimedia editing through the same four-criterion design cycle
- IBUnderstand that any AI-generated work included, even in part, in a piece of assessment is not considered the student's own unless it is clearly credited in the text and referenced in the bibliography
- IBFollow the IB Academic Integrity policy's guidance on the ethical use of AI tools (per its dedicated appendix) when producing assessed work
- England + CambridgeDevelop capability, creativity and knowledge in computer science, digital media and information technology (statutory for all KS4 pupils, not only GCSE candidates)
- CSTA (US)Evaluate licenses that limit or restrict use of computational artifacts when using resources such as libraries.
- CSTA (US)Explain the beneficial and harmful effects that intellectual property laws can have on innovation.
- OECD–EC (Europe)Purposefully integrate AI outputs with original ideas to build a comprehensive solution (Create 1, advanced).
- OECD–EC (Europe)Select and integrate outputs from multiple AI systems into a final product (Create 2, advanced).
- OECD–EC (Europe)Consider when attribution, permission, or avoiding AI use is the appropriate choice (Create 4, intermediate).
- OECD–EC (Europe)Assess the ethical implications and justify how an AI-assisted work maintains authenticity (Create 4, advanced).
- OECD–EC (Europe)Be aware that AI can help edit and process digital content (e.g. photo aging, text auto-suggestion), and know how to incorporate AI-edited or AI-generated content into one's own work while considering questions of credit (AI examples 7 and 38, competence 3.2).
- UNESCOInterrogate how automation of literature reviews, writing and artistic creation by AI may undermine human thinking and intellectual development, and identify concrete actions to protect themselves and peers from over-reliance on AI outputs (CG4.2.1.3).
- UNESCOCarry out responsible AI practices in compliance with ethical principles and locally applicable regulations, iteratively building and following a personal checklist of lawful and responsible practices (e.g. protecting privacy/copyright, marking AI-generated content, avoiding disinformation, hate speech, or sensitive content about identifiable individuals) (CG4.2.2.1).
- UNESCOGain an age-appropriate understanding of personal, legal and ethical responsibilities when using AI and reinforce self-disciplined behaviours around sensitive personal data, copyrighted materials, deepfake/identifiable-person imagery, and the spread of misinformation, disinformation or hate speech (CG4.2.2.2).
- UNESCOReflect on personal goals and a lifelong-learning, self-actualization attitude toward AI as it becomes pervasive in work, life and social practices (CG4.3.1.3).
- UNESCOCultivate social responsibilities to uphold 'ethics by design' in regulations on AI — evaluating how selected local AI regulations align with ethics-by-design, and drafting proposals to modify or create new regulations governing ethics by design (CG4.3.2.3).
- UNESCONurture a self-identity as a co-creator and 'driver' of the next generation of AI technology, critically analyse the long-term impacts of AI systems on social relations and behaviours, and discuss how regulations/policies should adapt to enhance AI governance (e.g. by engaging with local/online communities of AI co-creators) (CG4.3.4.3).
- SingaporeKeep digital knowledge current, including awareness of emerging technologies such as AI, as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital information management skills as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital communication, collaboration and participation skills as one of the nine core NDLP digital competencies.
- SingaporeApply the Find-Think-Apply-Create framework when locating, evaluating, using and creating with digital information (exact breakdown of the four components not published on the primary MOE page reviewed).
- SingaporeLearn to use AI: use AI tools safely and responsibly.
- SingaporeLearn beyond AI: develop distinctly human capabilities and dispositions that AI cannot replicate.
- SingaporeDevelop digital literacy and technological skills needed to thrive in an increasingly technology-driven world, as a central goal of the Masterplan.
- Korea, Japan, HKUnderstand the properties of information and media, and use information and information technology to discover and solve problems in the information society.
- Australia + NZApply digital literacy skills (investigating with, creating with, communicating and collaborating with, and managing/operating digital tools) across all learning areas, including recognising the role of AI-enabled tools in everyday digital systems
- Australia + NZStudents understand that generative AI tools should be used in ways that benefit all members of the school community and support wellbeing, not just individual convenience
- Australia + NZStudents learn that use of generative AI outputs (e.g. in assessable work) must remain open to challenge, retain human agency, and that the student/user remains accountable for decisions made with AI assistance (e.g. academic integrity implications)
- Australia + NZStudents understand that their privacy and personal data must be protected when using generative AI tools, and are taught not to input sensitive personal or others' data into such tools without appropriate safeguards
Unit 6: Deep Learning Projects 12 topics
Spark of Inspiration32 lessons· meets 47 requirements inUAECBSECISCE + statesIB+8
Lessons
- Spark of Inspiration
- Three Magic Questions
- Why One Small Idea Wins
- Finding Real Problems
- Great Projects Start Narrow
- Does It Need AI?
- Problem-Solution Match
- Why Problem-First Beats Tool-First
- AI Capabilities
- Superpower Sorting
- Reality Check
- Which Check Fails?
- The Idea Statement
- Fix the Idea Statement
- Choose Your Own Problem
- Scope Maya's Own Problem
- Who Needs It, and Does It Fit Our Values?
- Values Check: Maya's Garden App
- Your Workforce and Values Section
- Your Ideas First, Then the AI's
- Sort Maya's Ten Ideas
- Widen Your Ideas With AI
- Is This Merge Original?
- Merged Idea Statement
- Test an Idea Before Building
- Why Duolingo Starts With Specific Tasks
- Ideation Process
- Idea Checker
- Ideation Quiz
- Explain a Buildable AI Idea
- Ideation Vocab
- Ideation Champion!
Requirements this topic helps meet
- UAEImprove AI systems to match societal expectations
- UAEDesign AI solutions with systematic design processes
- CBSEIdentify and appreciate AI and describe its applications in daily life; relate, apply and reflect on Human-Machine Interaction across the three domains of AI - Data, Computer Vision, and Natural Language Processing; undergo assessment for AI-readiness skills (Unit 1: AI Reflection, Project Cycle and Ethics).
- CBSEIdentify the AI Project Cycle framework (Problem Scoping, Data Acquisition, Data Exploration, Modeling, Evaluation, Deployment); learn problem scoping and goal-setting for an AI project; identify stakeholders; brainstorm ethical issues involved in the problem selected; identify data requirements and reliable sources.
- CBSEComplete a Part D project/field visit/student portfolio related to Sustainable Development Goals, e.g. building an AI model with Teachable Machine or Machine Learning for Kids, or maintaining a portfolio of AI activities (4Ws canvas, system map, research on AI for SDGs).
- CBSEUnderstand the stages of the AI Project Cycle and the three AI domains with illustrations of practical applications; learn about ethical frameworks for AI and their categories, including Bioethics as a framework used in healthcare (Unit 1).
- CBSEEmpower students to create positive change through AI-driven social impact projects, and equip students with skills to develop AI solutions addressing societal challenges (course objectives).
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesApply the AI Project Framework/Cycle stages — problem scoping, data acquisition, data exploration, modelling and evaluation — to a use case (in brief)
- CISCE + statesApply the 4Ws problem canvas (Who, What, Where, Why) to scope a problem and write a problem statement; identify data sources/features/datasets including the Kaggle platform for data acquisition
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBGiven a design situation, identify a problem that needs to be solved and analyse the need for a solution
- IBConduct an inquiry into the nature of a design problem to inform the design brief
- IBWrite a detailed design specification that drives the development of a solution
- IBDevelop and present a proposed design solution derived from the specification
- IBDesign tests to evaluate a solution, carry out those tests, and objectively evaluate the solution's success
- IBIdentify areas where a design solution could be improved and explain the solution's impact on the client or target audience
- England + CambridgeYoung people should understand how AI works, its capabilities and limitations, and learn to use it effectively without becoming dependent on it; this should be delivered primarily through Computing rather than assumed to develop automatically
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Analyze the risks, benefits, and effectiveness of using AI for specific tasks (e.g., coding, brainstorming), including when AI is used to fully automate a process or is used with a human-in-the-loop approach.
- CSTA (US)Evaluate the design, motivation, outcomes, and potential impacts of AI systems using ethical design criteria and/or ethical frameworks.
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from identifying current uses of AI and how they have impacted society (K-2), to identifying how sectors of society such as manufacturing, retail, and healthcare are changing due to AI (3-5), to comparing AI-driven societal change to previous industrial revolutions (6-8), to predicting how a specific sector of society is likely to change in the short and intermediate term as a result of AI, referencing Amara's Law (9-12).
- AI4K12 (US)Progress from describing how AI can be used to solve a societal problem, e.g. Code.org's AI for Oceans classifying wildlife vs. debris (K-2), to designing a solution to a societal problem using AI technology, e.g. environmental protection, energy conservation (3-5), to researching a societal problem and describing how AI perception/reasoning/learning technologies could address it, referencing the UN Sustainable Development Goals (6-8), to evaluating an actual AI-for-social-good project in terms of the problem addressed and its real or potential impact (9-12).
- OECD–EC (Europe)Evaluate how well an AI use aligns with ethical principles and human values across trade-offs (Engage 7, advanced).
- OECD–EC (Europe)Purposefully integrate AI outputs with original ideas to build a comprehensive solution (Create 1, advanced).
- OECD–EC (Europe)Compare task complexity, the need for human judgment, and ethical implications when deciding whether to use AI (Manage 1, advanced).
- OECD–EC (Europe)Assess an AI system's strengths and limitations considering its purpose and potential impacts (Shape 1, advanced).
- OECD–EC (Europe)Design improvements to an AI system based on technical constraints, ethics, and user needs to promote societal benefit (Shape 4, advanced).
- OECD–EC (Europe)States that a foundational understanding of AI's technical functioning and the ability to recognise chances, limits and risks are indispensable for students to use AI responsibly and reflectively ('digital mündig'); calls on the Länder to integrate AI-related informatics competences and 'Bildung in der digitalen Welt' competences into curricula for all subjects and school types, without itself specifying discrete grade-level student outcomes.
- UNESCODiscuss age-appropriate real-world cases around six core AI ethical principles — do no harm, proportionality, non-discrimination, sustainability, human determination, and transparency/explainability — and their personal implications (CG4.1.2.2).
- UNESCOInternalise ethical principles on AI (human rights, data privacy, human agency, equity, inclusion, social justice, environmental sustainability) into reflective practices, e.g. advocating for inclusive AI interfaces and reporting discriminatory bias found in AI tools (CG4.1.2.3).
- UNESCOCritically examine, from legal, ethical and logical perspectives, whether AI should or should not be used for a given real-world challenge, and define the boundaries, goals and constraints of a problem before attempting to train an AI model (CG4.1.4.1).
- UNESCOPractise skills to identify and scope a problem that could be solved by building or training a new AI model, based on a simulation project (CG4.1.4.2).
- UNESCODevelop skills to assess an AI system's needs for data, algorithms, programming languages, software, and computing hardware, including feasibility given regulatory/ethical restrictions and cost (CG4.1.4.3).
- UNESCOApply an integral, ethics-by-design set of principles across the whole AI life cycle, e.g. simulating the role of a 'chief ethics officer' drafting a checklist of ethical criteria to audit each step of an AI system's design (CG4.3.2.1).
- UNESCOCustomize assessment methods/instruments to test self-crafted AI tools for robustness and ease of use, organise peer assessment and shared feedback, and build collaborative co-creator skills (CG4.3.3.3).
- SingaporeUnderstand AI's uses, risks and limitations, and apply ethical considerations (e.g. academic integrity, proper data handling) when using AI to support learning.
- SingaporeIdentify and explain AI's potential benefits, limitations and risks.
- Korea, Japan, HKJudge, based on scientific reasoning, the appropriate and effective use of information and information technology suited to purpose and situation, and consider what constitutes a desirable information society.
- ChinaStudents construct/build simple AI algorithm models and develop cross-disciplinary solutions using accumulated AI knowledge
The Data Hunt22 lessons· meets 13 requirements inCBSECISCE + statesEngland + CambridgeCSTA (US)+2
Lessons
- The Data Hunt
- Finding Datasets
- Why More Data Can Still Fail
- Data Quality 101
- Cleaning Data Changes Everything
- Quality Matcher
- Why the Test Set Stays Secret
- Data Cleaning
- Cleaning Sorter
- Clean It in Code
- Train-Test Split
- Split It Right
- Data Augmentation
- Does the Label Survive?
- Tune a Dataset Plan
- Why Spotify Needs Clean Feedback Data
- Prep Pipeline
- Data Doctor
- Data Prep Quiz
- Explain Good Dataset Prep
- Data Vocabulary
- Data Master!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Evaluate the data used to solve a problem, including its source(s) and whether privacy is protected, if/how the data has been processed, data quality (e.g., accuracy, reliability, validity), what the data represents, and biases.
- CSTA (US)Students build data fluency and examine the societal impacts of large-scale data collection used to train AI systems.
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Evaluate how dataset properties affect an AI system's performance and impact, factoring in data source and selection (Shape 3, advanced).
The AI Blueprint25 lessons· meets 17 requirements inADEKCBSECISCE + statesEngland + Cambridge+4
Lessons
- The AI Blueprint
- Layer Types
- Count the Weights
- Why the Wrong Architecture Struggles
- Famous Architectures
- Structure Creates Intelligence
- Architecture Match
- Why Simpler Often Wins First
- Model Anatomy
- Data Flow Sorter
- Shape Check
- Hyperparameters
- Chosen or Learned?
- Design Principles
- Tune a Model Blueprint
- How Gmail Matches Model to Task
- Design Process
- Architecture Judge
- Ethics by Design, Layer by Layer
- Which Build Step?
- Your Ethics Canvas
- Architecture Quiz
- Explain Architecture Fit
- Architecture Vocab
- Model Architect!
Requirements this topic helps meet
- ADEKDesign a scalable AI system architecture
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from modeling the use of a classification or search algorithm to solve a simple problem (K-2), to describing differences between decision-tree and neural-network classification algorithms (3-5), to comparing several algorithms that could solve a specific reasoning problem based on input data and decision complexity (6-8), to listing, for each of eight reasoning-problem types, an algorithm that could solve it (9-12).
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from analyzing a decision-tree-building game and how the learning algorithm adds/replaces nodes (3-5), to comparing how decision-tree learning (add a node) differs from neural-network learning (adjust every weight) (6-8), to describing how decision-tree, neural-net, regression, and reinforcement-learning algorithms each adjust their own type of internal representation (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- UNESCOAcquire methodological and technical skills to evaluate frameworks/components (e.g. TensorFlow, PyTorch, Scikit-learn) and configure a scalable, maintainable, reusable prototype AI architecture (data, algorithm, model and interface layers) for a defined problem, considering open-source options (CG4.2.4.1).
- UNESCOAcquire the interdisciplinary technical skills and project-management competencies needed to build a prototype AI system for a simple specific task (e.g. a chatbot), including leveraging computing resources, selecting/enhancing AI models (e.g. hyperparameter optimization), and balancing cost against computing needs and environmental impact (CG4.2.4.2).
- ChinaStudents build a three-dimensional 'technical principle → system architecture → social impact' way of reasoning about AI
- Australia + NZStudents understand that generative AI tools should be used in ways that benefit all members of the school community and support wellbeing, not just individual convenience
Teaching Your AI22 lessons· meets 10 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- Teaching Your AI
- The Training Loop
- Why Training Can Get Worse
- Loss Functions
- Compute the Loss
- A Curve Can Tell a Story
- Loss Matcher
- Why One Setting Can Change Everything
- Meet the Optimizers
- One Update Step
- Epochs & Batches
- Training Curves
- Diagnose the Curve
- Fighting Overfitting
- Tune a Training Run
- Why Tesla Tracks Validation Closely
- Training Pipeline
- Training Judge
- Training Quiz
- Explain Overfitting and Monitoring
- Training Vocab
- Training Master!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Build a Vision AI22 lessons· meets 11 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- Build a Vision AI
- CNN Architecture
- Run the Filter
- How Pixels Become Meaning
- Famous Datasets
- A CNN Learns in Stages
- Dataset Matcher
- Why Small Image Choices Matter
- Image Preprocessing
- Prep Sorter
- CNN Blueprint
- Trace the Shapes
- Training the CNN
- Diagnose the Model
- Tune an Image Classifier
- How Google Lens Reads Images Fast
- Project Pipeline
- CNN Judge
- Image Class Quiz
- Explain How a CNN Sees
- CNN Vocabulary
- Vision AI Builder!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from illustrating why training a classifier for a broad concept like 'dog' requires large, diverse image data by browsing datasets such as ImageNet or Coco (3-5), to illustrating how objects in images are segmented and labeled with bounding boxes to build an object-recognition training set (6-8), to evaluating a real AI training dataset's size, acquisition/labeling method, storage cost, and time-to-produce, including where bias can enter (9-12). K-2 is marked N/A.
AI That Reads22 lessons· meets 11 requirements inCBSECISCE + statesEngland + CambridgeAI4K12 (US)
Lessons
- AI That Reads
- Text vs Images
- Why Text Confuses Machines
- Text Preprocessing
- Run the Pipeline
- Meaning Lives Between Words
- Prep Matcher
- Why Padding Still Helps
- Word Embeddings
- Close in Meaning?
- Text Tasks Sorter
- Text Datasets
- Text Model Layers
- Trace the Model
- Tune a Text Pipeline
- How Gmail Detects Spam Fast
- Text Project Pipeline
- Text Classifier Judge
- Text Class Quiz
- Explain Text Preprocessing
- NLP Vocabulary
- NLP Builder!
Requirements this topic helps meet
- CBSEIdentify and appreciate AI and describe its applications in daily life; relate, apply and reflect on Human-Machine Interaction across the three domains of AI - Data, Computer Vision, and Natural Language Processing; undergo assessment for AI-readiness skills (Unit 1: AI Reflection, Project Cycle and Ethics).
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
Model Optimization22 lessons· meets 21 requirements inADEKCBSECISCE + statesIB+5
Lessons
- The Slow Model
- Optimization Goals
- Why Smaller Can Stay Smart
- Model Pruning
- Prune by Hand
- Efficiency Is a Design Skill
- Technique Match
- Why Profiling Comes First
- Quantization
- Quantize It
- Trade-Off Sort
- Knowledge Distillation
- Hardware Optimization
- Pick the Right Tool
- Balance an Optimization Plan
- Why Snapchat Needs Fast Vision Models
- Optimization Steps
- Spot Optimization
- Optimization Quiz
- Explain Model Optimization
- Vocab Cards
- Speed Champion!
Requirements this topic helps meet
- ADEKDesign a scalable AI system architecture
- ADEKFootprint / resource cost of AI
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBDesign tests to evaluate a solution, carry out those tests, and objectively evaluate the solution's success
- IBIdentify areas where a design solution could be improved and explain the solution's impact on the client or target audience
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Design ways to minimize negative environmental impacts of AI and communicate those ways to others.
- CSTA (US)Students examine security implications and environmental impacts of AI systems within Systems & Security.
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how design and deployment decisions affect an AI system's resource consumption (Engage 5, advanced).
- OECD–EC (Europe)Consider the ethical consequences of AI systems throughout their life-cycle, including environmental impact and societal impact such as platformisation of work and low-cost labelling labour (AI example 57, competence 4.4).
- UNESCOAcquire methodological and technical skills to evaluate frameworks/components (e.g. TensorFlow, PyTorch, Scikit-learn) and configure a scalable, maintainable, reusable prototype AI architecture (data, algorithm, model and interface layers) for a defined problem, considering open-source options (CG4.2.4.1).
- UNESCOAcquire the interdisciplinary technical skills and project-management competencies needed to build a prototype AI system for a simple specific task (e.g. a chatbot), including leveraging computing resources, selecting/enhancing AI models (e.g. hyperparameter optimization), and balancing cost against computing needs and environmental impact (CG4.2.4.2).
- UNESCOCustomize assessment methods/instruments to test self-crafted AI tools for robustness and ease of use, organise peer assessment and shared feedback, and build collaborative co-creator skills (CG4.3.3.3).
Deployment Basics25 lessons· meets 28 requirements inADEKCBSECISCE + statesEngland + Cambridge+6
Lessons
- Model Goes Live!
- Where to Deploy
- Cloud, Edge or Browser?
- Why a Great Model Still Isn't Useful
- What is an API?
- Read an API Call
- Serving AI Is a Separate Skill
- Component Match
- Why Monitoring Never Ends
- Containers (Docker)
- Environment Sort
- Scaling Secrets
- Production Monitoring
- Read the Dashboard
- Balance a Deployment Plan
- A Front Door for Your Model
- Build Your Model's Web Page
- What Did Your Tests Show?
- How Netflix Serves Recommendations
- Deployment Steps
- Spot Red Flags
- Deployment Quiz
- Explain AI Deployment
- Vocab Cards
- Deploy Master!
Requirements this topic helps meet
- ADEKDesign a scalable AI system architecture
- ADEKFootprint / resource cost of AI
- CBSEIdentify the AI Project Cycle framework (Problem Scoping, Data Acquisition, Data Exploration, Modeling, Evaluation, Deployment); learn problem scoping and goal-setting for an AI project; identify stakeholders; brainstorm ethical issues involved in the problem selected; identify data requirements and reliable sources.
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesApply the AI Project Framework/Cycle stages — problem scoping, data acquisition, data exploration, modelling and evaluation — to a use case (in brief)
- CISCE + statesApply the 4Ws problem canvas (Who, What, Where, Why) to scope a problem and write a problem statement; identify data sources/features/datasets including the Kaggle platform for data acquisition
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Design ways to minimize negative environmental impacts of AI and communicate those ways to others.
- CSTA (US)Students examine security implications and environmental impacts of AI systems within Systems & Security.
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- OECD–EC (Europe)Analyse how design and deployment decisions affect an AI system's resource consumption (Engage 5, advanced).
- OECD–EC (Europe)Establish checkpoints, monitor AI use against criteria, and adjust roles throughout a problem-solving process (Manage 4, advanced).
- OECD–EC (Europe)Consider the ethical consequences of AI systems throughout their life-cycle, including environmental impact and societal impact such as platformisation of work and low-cost labelling labour (AI example 57, competence 4.4).
- UNESCOUnderstand the necessity of exercising sufficient human control over AI, including the consequences when human oversight fails or is absent (CG4.1.1.2).
- UNESCOPractise skills to identify and scope a problem that could be solved by building or training a new AI model, based on a simulation project (CG4.1.4.2).
- UNESCODevelop skills to assess an AI system's needs for data, algorithms, programming languages, software, and computing hardware, including feasibility given regulatory/ethical restrictions and cost (CG4.1.4.3).
- UNESCOAcquire methodological and technical skills to evaluate frameworks/components (e.g. TensorFlow, PyTorch, Scikit-learn) and configure a scalable, maintainable, reusable prototype AI architecture (data, algorithm, model and interface layers) for a defined problem, considering open-source options (CG4.2.4.1).
- UNESCOAcquire the interdisciplinary technical skills and project-management competencies needed to build a prototype AI system for a simple specific task (e.g. a chatbot), including leveraging computing resources, selecting/enhancing AI models (e.g. hyperparameter optimization), and balancing cost against computing needs and environmental impact (CG4.2.4.2).
- UNESCOBuild technical and social-responsibility skills to make decisions on optimizing, reconfiguring, or shutting down an AI system based on test results and user feedback (three scenarios: optimize datasets/model/interface; reconfigure problem-scoping/architecture; or shut down and remediate where the system violates human rights or harms vulnerable groups) (CG4.3.4.2).
- Korea, Japan, HKStudents and teachers should treat generative AI output as one reference among others, always keeping human judgement and responsibility final, and use the information only after checking it against the qualities/abilities set out in the curriculum guidelines.
- Australia + NZDevelop, modify and implement software (e.g. webpages using HTML for structure and CSS for presentation) that address defined needs, including validation and testing
Project Documentation22 lessons· meets 26 requirements inCBSECISCE + statesIBEngland + Cambridge+5
Lessons
- The Lost Project
- Why Document?
- Why Good Projects Get Lost
- The README
- Fix This README
- Writing Saves Engineering Time
- Doc Section Match
- Why README First Still Works
- Model Cards
- Audience Sort
- Code Comments
- Which Comment Helps?
- Experiment Tracking
- Read the Run Log
- Tune a Documentation Plan
- How Hugging Face Model Cards Help
- Documentation Steps
- Spot Good Docs
- Documentation Quiz
- Explain Why Documentation Matters
- Vocab Cards
- Doc Champion!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CBSEComplete Unit 3: Information and Communication Technology Skills (Level I-IV progressing Class IX to XII), covering topics such as email introduction and account creation, and networking basics at Class IX per secondary sources.
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBUse and apply technology effectively to access, process, and communicate information, and to model and create solutions to problems
- England + CambridgeDevelop capability, creativity and knowledge in computer science, digital media and information technology (statutory for all KS4 pupils, not only GCSE candidates)
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Evaluate and refine computational artifacts to make them more usable and accessible.
- CSTA (US)Design and develop computational artifacts working in team roles using collaborative tools.
- CSTA (US)Use tools and methods for collaboration on a project to increase connectivity of people in different cultures and career fields.
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- SingaporeKeep digital knowledge current, including awareness of emerging technologies such as AI, as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital information management skills as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital communication, collaboration and participation skills as one of the nine core NDLP digital competencies.
- SingaporeApply the Find-Think-Apply-Create framework when locating, evaluating, using and creating with digital information (exact breakdown of the four components not published on the primary MOE page reviewed).
- SingaporeDevelop digital literacy and technological skills needed to thrive in an increasingly technology-driven world, as a central goal of the Masterplan.
- Korea, Japan, HKUnderstand the properties of information and media, and use information and information technology to discover and solve problems in the information society.
- Korea, Japan, HKUnderstand and apply the concepts and methods of information design (including universal design, usability and accessibility) needed for effective communication with a given audience, and evaluate/improve created content.
- Australia + NZDesign the user experience of a digital system generating or processing data, and evaluate the suitability of user interfaces for defined purposes and end-users, applying accessibility standards for diverse audiences
- Australia + NZPlan and manage projects individually and collaboratively, including using project management tools/strategies to sequence production activities
- Australia + NZApply digital literacy skills (investigating with, creating with, communicating and collaborating with, and managing/operating digital tools) across all learning areas, including recognising the role of AI-enabled tools in everyday digital systems
Presentation and Demo25 lessons· meets 24 requirements inUAEADEKCBSECISCE + states+7
Lessons
- Demo Day Arrives!
- Demo Structure
- Why a Strong Project Can Still Fall Flat
- Storytelling Tips
- Demos Translate Complexity
- Demo Part Match
- Why Backup Plans Build Confidence
- Live Demo Magic
- Honest Demo or Cherry-Pick?
- Demo Prep Sort
- Q&A Mastery
- Answer the Judge
- Slide Design
- Slide, Say or Skip?
- Tune a Demo Plan
- How Apple Keynotes Demo AI Clearly
- Presentation Flow
- Spot Good Demos
- Presentation Quiz
- Deliver It: Record Your 3-Minute Demo
- Score Your Own Talk
- Your Self-Assessment
- Explain What Makes a Demo Work
- Vocab Cards
- Presenter Pro!
Requirements this topic helps meet
- UAEDesign AI solutions with systematic design processes
- ADEKBuild a data story (data + visuals + narrative for an audience)
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CBSEComplete Unit 3: Information and Communication Technology Skills (Level I-IV progressing Class IX to XII), covering topics such as email introduction and account creation, and networking basics at Class IX per secondary sources.
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- IBUse and apply technology effectively to access, process, and communicate information, and to model and create solutions to problems
- IBWhere a school offers digital design (rather than or alongside product design), demonstrate digital-tool competencies such as prototyping, graphic/web design, and multimedia editing through the same four-criterion design cycle
- England + CambridgeDevelop capability, creativity and knowledge in computer science, digital media and information technology (statutory for all KS4 pupils, not only GCSE candidates)
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- SingaporeKeep digital knowledge current, including awareness of emerging technologies such as AI, as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital information management skills as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital communication, collaboration and participation skills as one of the nine core NDLP digital competencies.
- SingaporeApply the Find-Think-Apply-Create framework when locating, evaluating, using and creating with digital information (exact breakdown of the four components not published on the primary MOE page reviewed).
- SingaporeDevelop digital literacy and technological skills needed to thrive in an increasingly technology-driven world, as a central goal of the Masterplan.
- Korea, Japan, HKUnderstand the properties of information and media, and use information and information technology to discover and solve problems in the information society.
- ChinaStudents construct/build simple AI algorithm models and develop cross-disciplinary solutions using accumulated AI knowledge
- Australia + NZApply digital literacy skills (investigating with, creating with, communicating and collaborating with, and managing/operating digital tools) across all learning areas, including recognising the role of AI-enabled tools in everyday digital systems
Deep Learning Portfolio22 lessons· meets 20 requirements inCBSECISCE + statesEngland + CambridgeCSTA (US)+4
Lessons
- Your AI Gallery
- Where to Host
- Why Do Some Portfolios Get Noticed Fast?
- Project Essentials
- README: Strong or Weak?
- A Portfolio Can Compress Years of Learning
- Portfolio Match
- Story and Evidence Beat a Long Project List
- Project Showcase
- Project Sort
- Credit Where It's Due
- GitHub Mastery
- Your Story
- Share or Keep Private?
- Build Steps
- Tune a Portfolio Showcase
- How GitHub Profiles Showcase Serious Builders
- Spot Best Practices
- Portfolio Quiz
- Explain What Makes a Portfolio Memorable
- Vocab Cards
- Portfolio Ready!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- England + CambridgeUnderstand how changes in technology affect safety, including new ways to protect online privacy and identity, and how to identify and report a range of concerns
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Evaluate licenses that limit or restrict use of computational artifacts when using resources such as libraries.
- CSTA (US)Design and develop computational artifacts working in team roles using collaborative tools.
- CSTA (US)Use tools and methods for collaboration on a project to increase connectivity of people in different cultures and career fields.
- CSTA (US)Explain the beneficial and harmful effects that intellectual property laws can have on innovation.
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- SingaporeDevelop digital safety and security competencies as one of the nine core NDLP digital competencies.
- SingaporeDevelop digital responsibility as one of the nine core NDLP digital competencies.
- Korea, Japan, HKUnderstand laws/regulations on information, the importance of information security, and individual responsibility and information morals in the information society.
- Australia + NZPlan and manage projects individually and collaboratively, including using project management tools/strategies to sequence production activities
- Australia + NZStudents understand that their privacy and personal data must be protected when using generative AI tools, and are taught not to input sensitive personal or others' data into such tools without appropriate safeguards
Year Review & Grade 11 Preview10 lessons· meets 16 requirements inCBSECISCE + statesEngland + CambridgeCSTA (US)+2
Lessons
- What a Journey!
- Your 6 Units
- Worked Example: One Project, Four Units
- Your Project Plan, In Order
- Pick the Right Tool
- Will It Fit on a Phone?
- Claim or Overclaim?
- A Crop-Disease Model That Runs in the Field
- Final Quiz
- Grade 10 Complete!
Requirements this topic helps meet
- CBSEExplain the basic structure and components of a neural network; identify different types of neural networks and their applications; understand machine learning and neural networks through hands-on projects (e.g. TensorFlow Celsius-to-Fahrenheit model, TensorFlow Playground classification) (Unit 6: Understanding Neural Networks).
- CISCE + statesExplain Artificial Neural Networks (ANN) — neurons, input/hidden/output layers, forward and backward propagation, multi-layer perceptron (MLP) — using an example such as handwritten-digit recognition
- CISCE + statesBuild, train and evaluate a neural network using Keras/TensorFlow (fit(), epochs, batch size) and evaluate model performance using accuracy, precision and recall, visualized with Matplotlib
- CISCE + statesImplement a simple image classifier distinguishing categories of objects (e.g. cats vs dogs, fruits vs vegetables) and build a facial-recognition application that can associate identified faces with known individuals
- CISCE + statesTrack the movement of objects across a video sequence over time (e.g. a ball in sports footage, vehicles in traffic surveillance)
- England + CambridgeShow understanding of how artificial neural networks have helped with machine learning
- England + CambridgeShow understanding of back propagation of errors and regression methods in machine learning
- CSTA (US)Debate what differences do or should exist between human and artificial intelligence, sentience, consciousness, rights, and responsibilities.
- AI4K12 (US)Progress from identifying a problem as a classification problem or a search problem (K-2), to categorizing problems as classification vs. search using examples like Teachable Machine (3-5), to categorizing problems as classification, prediction, combinatorial search, or sequential decision problems (6-8), to categorizing real-world problems across eight reasoning-problem types including heuristic search, adversarial search, logical deduction, and statistical inference (9-12).
- AI4K12 (US)Progress from demonstrating, with instructor help, how to train a computer to recognize something using a tool like Teachable Machine (K-2), to training a classification model and examining its accuracy on new inputs (3-5), to training and evaluating a classification or prediction model on a tabular dataset (6-8), to using supervised or unsupervised learning on real-world data and quantitatively evaluating the results on a nontrivial test set (9-12).
- AI4K12 (US)Progress from explaining how reinforcement learning lets a computer learn from trial-and-error experience with a reward signal, e.g. a grid-world navigation demo (3-5), to explaining the difference between supervised learning (told the correct output) and reinforcement learning (told only how well it's doing) (6-8), to selecting the appropriate learning-algorithm family (supervised, unsupervised, or reinforcement) for a given reasoning problem (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from illustrating that a 1-3 neuron network is a function that computes weighted-sum/threshold outputs (3-5), to illustrating the layered structure (input/hidden/output) of a neural network and its transfer functions (6-8), to describing feed-forward, 2D convolutional, recurrent, and generative adversarial network architectures and their typical uses, e.g. deepfakes (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating how weights are assigned in a neural network to produce a desired input/output behavior (3-5), to demonstrating a learning rule (e.g. perceptron rule) that adjusts weights of a one-layer network in response to errors (6-8), to training a multilayer neural network with the backpropagation algorithm and describing how weights and hidden-unit outputs change during learning (9-12). K-2 is marked N/A.
- AI4K12 (US)Progress from demonstrating ways intelligent agents don't understand things the way people do, e.g. no frustration at repeated questions (K-2), to demonstrating what a 'broad AI'/general-AI system could do that current narrow AI cannot (3-5), to defining criteria for consciousness (self-awareness, personal history) and evaluating AI systems or fictional AI characters against them (6-8), to debating alternative perspectives on human vs. artificial intelligence, including the Turing test and artificial superintelligence (9-12).
- UNESCODevelop an age-appropriate understanding of the three types of AI algorithms — supervised learning, unsupervised learning, and reinforcement learning — and how the data behind them is acquired and labelled (CG4.1.3.2).
- UNESCOReflect on personal goals and a lifelong-learning, self-actualization attitude toward AI as it becomes pervasive in work, life and social practices (CG4.3.1.3).
Bring the course to your school or region
Schools, school groups, resellers and partners: we will walk you through the alignment for your board and the lessons behind it.
Our own reading of our own course, ratings reviewed 22 October 2026. It is not an endorsement, certification or accreditation by any authority named here, and it makes no claim about learning results.