Neural Network Playground · Virtual Labs · AquinTutor

← LABS Neural Network Playground REAL BACKPROP · FROM SCRATCH

Dataset

Noise 0.15 Train / test split 70%

Features (input)

Training

Learning rate 0.03 Activation L2 regularisation 0 Batch size 16

Live metrics

Epoch0

Train loss—

Test loss—

Train acc—

Test acc—

Parameters—

First-layer neurons

Each thumbnail is what one hidden neuron has learned to detect — its own decision surface. The output is a learned combination of them. This is a genuine MLP: forward pass, cross-entropy loss, and backpropagation are all computed here, in the browser, per step.