How a neural network learns
The short answer
A neural network learns by guessing, measuring its mistake and nudging its connection weights, over and over, until its answers come out right.
0:28Technology & AIIntermediateneural networksmachine learningAI
Chapters and transcript
- 0:00
It starts by guessing
A new network’s connections are random, so its first answers are mostly wrong. Shown a handwritten 7, it guesses 1.
- 0:07
Every connection has a weight
Signals flow through the layers from left to right. Each connection’s weight sets how strongly it passes the signal on.
- 0:14
Measure the mistake, adjust the weights
The network’s answer is compared with the right one. The error is traced back through the layers, and every weight moves a little to shrink it.
- 0:21
Repeat, thousands of times
With every example the answers improve, until the weights capture the pattern. The learning curve in the video is an illustration, not measured data.