One-shot Learning using Triplet Network with kNN classifier

Abstract

We propose a triplet network with a kNN classifier for the problem of one-shot learning, in which we predict the query images by given single example of each class. Our triplet network learns a mapping from sample images to the Euclidean space. Then we apply kNN classifier on the embeddings generated by the triplet network to classify the query sample. Our method can improve the performance of one-shot classification with data augmentation by processing the images. Our experiments on different datasets which are based on MNIST dataset demonstrate that our approach provides a effective way for one-shot learning problems.