Neural Network Playground

Design your own Multi-Layer Perceptron (MLP) to solve non-linear classification problems.

Architecture

Training Parameters

Learning Rate: 0.03

Data Presets

Epoch: 0

Loss: 0.0000

Parameters: 0

Hidden Layers: Allow the network to fold and warp the input space to separate complex classes.
Activation (Tanh/ReLU): The source of non-linearity. Without it, deep networks are just linear models!
Overfitting: Too many neurons + too few points = a jittery boundary that memorizes noise.