See the system
Make the important pieces and their relationships visible.
Interactive course · intro
Neural Networks
Build and train a neural network from first principles, then diagnose optimization, overfitting, and generalization.
- Length
- 74 minutes
- Lessons
- 10 lessons
- Format
- Interactive
INTERACTIVE COURSE · 10 lessonsNeural Networks
For people building products with AI who want a practical mental model they can use.
WHAT CHANGES
Turn a complex system into something you can reason about.
For people building products with AI who want a practical mental model they can use.
Make a decision
Use the model inside a realistic product or technical trade-off.
Explain the why
Leave with language you can use with product, design, and engineering teammates.
COURSE CURRICULUM
Short lessons. One connected mental model.
Each lesson makes one part of the system visible, then asks you to make a decision inside it.
ai tools
- 01INTERACTIVE LESSON5 MIN
A Neuron Makes a Guess
A weight turns an input into a guess, and the signed prediction gap shows which side of the target it lands on.
- 02INTERACTIVE LESSON7 MIN
A Neuron Makes a Decision
Turn a weighted score into a yes-or-no decision, then find the pattern one straight boundary cannot learn.
- 03INTERACTIVE LESSON8 MIN
Inside an MLP Neuron
Control the weights, bias, and activation that turn several inputs into one output.
- 04INTERACTIVE LESSON8 MIN
Build a Small Neural Network
Combine hidden neurons into a network that solves XOR, then see why choosing every setting by hand does not scale.
- 05INTERACTIVE LESSON7 MIN
Follow the Loss Downhill
Turn prediction error into a loss curve, then use its slope to move a weight toward a better setting.
- 06INTERACTIVE LESSON8 MIN
Send the Gradient Backward
Use the chain rule to compute each weight’s loss gradient, then expose why gradients can vanish in deep networks.
- 07INTERACTIVE LESSON8 MIN
Choose a Learning Rate
Tune the size of each training update, then diagnose slow, unstable, and healthy loss curves.
- 08INTERACTIVE LESSON8 MIN
Train the XOR Network
Assemble the complete learning loop, then watch one network’s weights, predictions, and loss improve together.
- 09INTERACTIVE LESSON8 MIN
Generalize Beyond the Training Set
Balance model capacity against held-out performance, then regularize an overfit neural network.
- 10INTERACTIVE LESSON7 MIN
Backprop in Real Systems
Match backpropagation’s mechanics to image models, language models, and automatic differentiation.
START FREE
Build a useful mental model in 74 minutes.
Try the first interactive lesson without an account or a credit card.