Neural network playground
Train a small network live in your browser. Press play and watch gradient descent bend the boundary between the two classes.
Epoch 0 · train 0.509 · test 0.496
Learning rate
Activation
Regularization
Regularization rate
Data
Inputs
2 hidden layers · 25 parameters
4
2
Each square shows what one neuron responds to across the plane. Line thickness is the size of a weight; colour is its sign.
Output
Test accuracy
68%
Epoch
0
Training lossTest loss
New to this? The lesson How a network learns walks through loss, gradient descent, learning rates and overfitting first. Inspired by TensorFlow Playground (Daniel Smilkov and Shan Carter); the network here is written from scratch and runs entirely on your device.