How a neural network learns

A guided example. Follow one cat image through a small deep network: forward pass, error, backpropagation, weight update, and watch the prediction go from wrong to correct as the loss falls.

GUIDED SIMULATION

One cat image moving through the network

3 inputs → 4 → 4 → 3 → 1 output
Training lossloss after each epoch (lower is better)
0Epoch
Cat probability
1Correct target
Loss
Prediction

Teaching simplification: this view follows one example so the idea stays visible. Real training uses mini-batches of many examples, averages their gradients, then updates the shared weights, repeated over many epochs.