Playground · Neural Net
Train a neural network, live
Pick a dataset (or click to place your own points), then hit Train and watch the network learn the boundary between the two classes. This is a real multi-layer perceptron with back-propagation — written from scratch in plain JavaScript, no libraries — running right in your browser.
Click the canvas to add a point of the selected class.
Dataset
Add points as
0
epoch
—
loss
—
accuracy
Architecture: 2 → 8 → 8 → 1, tanh hidden layers, sigmoid output, trained with full-batch gradient descent on binary cross-entropy. The background shading is the network's confidence across the whole plane. Curious how this works? See the 3Blue1Brown series on our Learn shelf.