backprop.sim // tiny network

INSTRUCTIONS
SAMPLE 1/1  ·  EPOCH 000  ·  LOSS —

How a neural network is trained, on a very simple example: forward pass → loss → backward pass (chain rule) → gradient descent update

FORWARD · activations BACKWARD · gradients
NETWORK GRAPH
INPUTS
DATA
INPUT x=1.5 w₁=0.80 ∂L/∂w₁= INPUT_1 x₁=1.5 INPUT_2 x₂=−1.0 w₁=0.80 w₂=0.60 ∂L/∂w₁= ∂L/∂w₂= NEURON_1 z₁=— a₁=— NEURON_2 z₂=— ŷ=— LOSS y=0.8 L=— SAMPLE 1 / 2 w₂=−0.50 ∂L/∂w₂=
[1] FWD in→n1 [2] FWD a₁→n2 [3] LOSS [4] BWD ∂L/∂w₂ [5] BWD ∂L/∂w₁ [6] UPDATE
epochs
TRACE // ONE PASS

MEAN LOSS / EPOCH
0.30 0

EPOCH 0 — STEP walks through each stage; RUN executes N epochs (fast); AUTO loops until paused.

backprop.sim // Developed by Armando Teixeira-Pinto and Claude Fable 5