Using the trained model

Load the learned parameters into the matching network. Give it a new example. The network produces an output, such as cat or dog.

This is called .

Training changes the parameters. Inference uses them.

Think of a student taking a test. During practice, an answer key exposes mistakes and the student adjusts. During the test, there is no answer key. The student uses what practice left behind and gives an answer.

During inference, a new input goes in and an output comes out. Nothing travels backward to correct that answer, so the learned parameters stay the same.