Neural Playground

Watch a network learn.

Draw examples from two classes and train a small neural network. Its decision boundary changes in front of you.

FR / LAB

Preparing the experiment…

Preparing controls.

01 / THE GESTURE

How to explore

Choose class A or B and add points to the plane. Press Start training. Keyboard-accessible controls can also add examples.

02 / THE MODEL

What you are observing

A 2 → 8 → 8 → 1 network with tanh and sigmoid activations, gradient descent and cross-entropy. Loss and accuracy describe the training examples, not unseen data. An educational classifier, not a language model.