Science & TechHow it works
How a neural network learns from its mistakes.
A neural network is a mesh of computing nodes linked by millions to billions of numbers, the weights. To learn, it takes an example, guesses an answer and compares it with the correct one. The error is passed backwards through the network, and each weight nudges a tiny step in the better direction. Repeated millions of times.
Useful: The network understands nothing; it has only found weights that keep the error small.
Source: en.wikipedia.org