Saving the settings

Training may take thousands or millions of correction steps. Each one nudges the parameters.

Imagine a machine covered in adjustable dials. Training turns them. Saving records the final position of every dial so the settings can be restored later.

In a neural network, a program saves the learned parameters in one file or several.

The files do not contain readable rules such as "pointed ears mean cat." They contain the numerical settings that training produced.

Training finds the settings. Saving makes them reusable.