Neural Network Backpropagation

Forward pass · chain rule gradient flow · weight updates · layer activations

Learning Rate: 0.10

Target Output: 0.80

Epoch: 0 | Loss (MSE): — | Phase: idle
Gradient magnitudes shown as edge thickness. Red = large gradient, blue = small. Chain rule: δL/δw = δL/δa · δa/δz · δz/δw.