How a model measures error

Why measure loss?

With random weights, a model is unlikely to produce the behavior we want, so it needs to be trained. Before training can happen, we need a way to measure how wrong its outputs are. That measurement is called loss.

The intended output—the number we want—is called the target. The demo lets you change five outputs and compare each one with its target.

What you will try

First compare an output with its target. Then square the error and add the five squared losses. Training uses this measurement to guide changes; we will explore that in the next meeting.