Watch a Neural Net Learn
Train a small network in your browser and watch its decision boundary and weights change at every step.
Coming soon
This tool is being built. It will run right here, in your browser.
How it works
Method
A small network trains on a 2D dataset you choose; each training step redraws the boundary, the loss curve and the weights.
Data
Generated 2D datasets (spirals, rings, clusters) and a handwritten-digit set for the bigger model.
Model
A multilayer perceptron trained with gradient descent, with adjustable layers, activations and learning rate.
Build
Hand-written training loop in TypeScript so every number on screen is the real one.