Hands-On One-shot Learning with Python

Training procedure

When it comes to training architecture, matching networks follow a certain technique: they try to replicate test conditions during the training phase. In simpler terms, as we have learned in the previous section, matching networks sample label sets from the training data, and later they generate a support set and a batch set from the same label set. After data preprocessing, matching networks learn their parameters by training the model to minimize the error by taking support sets as training sets, and batch sets as test sets. This training procedure of taking a support set as the training set and a batch set as the test set enables matching networks to replicate the test conditions.

In the next section, we will go through ...