Back to Playgrounds

Backpropagation Playground

Watch the math unfold step by step. Forward pass computes the prediction, backward pass computes the blame, weight update fixes the mistake.

2-input → 3-hidden (sigmoid) → 1-output | Input: [1, 0.5] | Target: 1 | LR: 0.5
x₁x₂h₁h₂h₃outInputHiddenOutput
Press Step or Auto-Play to begin
Ready

Weight Table

WeightValue
w₁₁0.1500
w₁₂0.2000
b₁0.3500
w₂₁0.2500
w₂₂0.3000
b₂0.3500
w₃₁0.1000
w₃₂0.1500
b₃0.3000
w₄₁0.4000
w₄₂0.4500
w₄₃0.5000
b₄0.6000
Forward Pass
Backward Pass
Weight Update
Active

All computation runs in your browser using plain JavaScript math. No frameworks, no GPU — just the raw forward pass, chain rule, and gradient descent that powers every neural network.