How a neural network learns
HE · 45 minutes · Explain forward pass, loss and backpropagation.
Answer by: Type. You can change method or take a break at any time.
Page 1 of 30
Our learning goal
Explain forward pass, loss and backpropagation.
Your turn: Choose one thing you would like to understand by the end.
Offline speech needs an installed English device voice.
My lesson notes & project
Full teaching script
- Today: How a neural network learns. Our goal: Explain forward pass, loss and backpropagation.
- Take one idea at a time. Pause after each sentence; ask for repetition whenever you need it.
- A network is layers of numbers. Each neuron multiplies its inputs by weights, adds a bias and passes the result through a non-linear function so the whole thing can bend, not just draw straight lines. The forward pass produces a prediction; the loss function scores how wrong it is. Backpropagation walks that error backwards using the chain rule, working out how much each weight contributed, and gradient descent nudges every weight a little downhill. Repeat over millions of examples and the weights become a compressed theory of the data. Overfitting is memorising the training set instead — held-out validation data is the honesty check.
- Together: choose one example from the explanation and show how it works. What do you notice?
- Your turn: Hand-compute one forward pass and one weight update for a 2-input, 1-output neuron.
- Reflect: what made sense, and what would you like explained differently?
Watch and learn
Videos come only from approved education channels and need internet. Use the arrows in the player for more videos from the same channel.