Select a preset and press Play to start training.

How It Works

A feedforward neural network learns by repeatedly adjusting its weights to minimize prediction error. Each training step has two phases:

Forward pass — input values propagate through the network layer by layer. Each neuron computes a weighted sum of its inputs, adds a bias, and applies an activation function.

Backward pass — the error (loss) at the output is propagated backward using the chain rule. Each weight receives a gradient indicating how much it contributed to the error.

Weight update — weights are nudged in the opposite direction of their gradient, scaled by the learning rate. Over many iterations the network converges to a solution.

a⁽ˡ⁾ = σ( W⁽ˡ⁾ · a⁽ˡ⁻¹⁾ + b⁽ˡ⁾ )

The loss function is binary cross-entropy (BCE), which measures divergence between predicted probabilities and binary targets. Best suited for classification tasks with sigmoid outputs. The learning rate controls how large each weight update step is — too high and training becomes unstable, too low and it converges slowly.

Dataset — XOR

#x1x2target
10.000.000.00
20.001.001.00
31.000.001.00
41.001.000.00

Presets

Controls

800ms
FastSlow
0.50

Training Info

Epoch

0

Sample

1/4

Loss

0.0000

Phase

Ready

Loss History

Train to see loss over time.

Network

2 → 3 → 1