How a Neural Network Learns: Weighted Sums, Errors and Tiny Nudges

A neural network is a pile of weighted sums with a squash after each one. It learns by comparing its output to the right answer and nudging every weight a little in the direction that reduces the error. This video shows a real, small network learning in front of you.

You'll see the forward pass, the error, the backward wave of corrections, and the weights settling into a pattern.

In this video

• Neurons as weighted sums with an activation
• Forward pass and the error
• Backpropagation as a wave of small nudges
• Why it needs many examples and many passes