43. Backpropagation Intuition (No Heavy Maths)
43.2 Forward Then Backward Story
Why. Students remember stories better than matrix calculus.
What. Imagine three friends in a line passing a message (forward). The last friend sees the mistake and whispers how much each earlier friend contributed (backward).
How.
Narrate one training step
- Layer A → Layer B → output.
- Loss says “too high by 0.4.”
- Output layer weights get the first blame.
- Hidden layer weights get scaled blame through the chain.
# Pseudocode — real API in Chapters 44–45
# pred = model(x)
# loss = criterion(pred, y)
# optimizer.zero_grad()
# loss.backward()
# optimizer.step()
print("Remember the four lines; details next.")