Short course · intro

Neural Networks

Build a neural network from its smallest piece, one neuron at a time.

0 of 9 read · 9 parts · ~51 min

1. 1 What one neuron computes A neuron multiplies each input by a weight, sums the results, and adds a bias. 4 min
2. 2 The activation function A neuron squashes its weighted sum through a nonlinear function, which is what lets stacked neurons do more than a straight line. 5 min
3. 3 From a neuron to a layer A layer is many neurons reading the same inputs at once, written compactly as one matrix equation. 5 min
4. 4 Stacking layers Feeding one layer's output into the next builds a deep network, and evaluating them in order is the forward pass. 5 min
5. 5 What the network represents A single neuron splits the input with a straight line; stacking layers with nonlinear activations bends that boundary into curves. 6 min
6. 6 Measuring how wrong the network is A loss function turns the gap between prediction and truth into a single number to minimize. 5 min
7. 7 Gradient descent Shifting each weight a small step against the slope of the loss is how a network learns. 5 min
8. 8 Backpropagation The chain rule computes each weight's slope as the product of the local slopes along the path to the loss. 6 min
9. 9 The training loop Forward pass, loss, gradients, update, repeated over the data, is how a network learns its weights. 6 min
10. 10 Course quiz Six questions checking the chain from one neuron to a trained network. 4 min