Playground experiment

MLP Decision Boundary Lab

Build a tiny neural net, switch activation functions, and watch its decision surface bend around a 2D dataset as training updates the weights.

Back to Playground

Decision surface behind the points blue/red regions show the current class probability

MLP controls

Choose a dataset, then train the network.

Loss --

Accuracy --

Updates 0

Dataset

Activation

Hidden layers 2 Neurons/layer 6 Learning rate 0.030 Dataset noise 8% Steps/tick 10

Tanh bends smoothly around the origin, which works well for centered toy data.