Hands-On Meta Learning with Python

Performing classification using prototypical networks

Now, we will see how to use prototypical networks to perform a classification task. We use an omniglot dataset for performing classification. This dataset comprises 1,623 handwritten characters from 50 different alphabets, and each character has 20 different examples written by different people. Since we want our network to learn from data, we train it in the same way. We sample five examples from each class and use that as our support set. We learn the embeddings of our support set using a sequence of four convolution blocks as our encoder and build the class prototype. Similarly, we sample five examples from each class for our query set, learn the query set embeddings, and predict ...