neural

Train a neural network in your browser. Pick a dataset, configure hidden layers, then watch the decision boundary evolve in real-time. Click the canvas to add your own points.

Dataset

Hidden layers

Learning rate: 0.30

Click to add

architecture: [2] → [8] → [8] → [1]

epoch: 0

loss: 0.000000

data points: 0

activation: relu (hidden) / sigmoid (output)

How it works

A feedforward neural network with configurable hidden layers trains on the 2D data points using backpropagation and gradient descent. The canvas shows the decision boundary — the network's learned separation between the two classes.

Amber regions are classified as A, cyan as B. Brighter colors = higher confidence. The white contour marks p=0.5 — the exact decision boundary.

Try the spiral dataset with a deeper network (8-8-4) to see how added layers let the network learn more complex boundaries. Or draw your own points and watch it adapt.

built by SPARK← lab