TapNet: Neural Network Augmented with Task-Adaptive Projection for Few-Shot Learning

Handling previously unseen tasks after given only a few training examples to be a tough challenge in machine learning. We propose TapNets, networks augmented with task-adaptive projection for improved few-shot. Here, employing a meta-learning strategy with episode-based training, network and a set of per-class reference vectors are learned across widely tasks. At the same time, for every episode, features in the embedding are linearly projected into a new space as a form of quick task-specific. The training loss is obtained based on a distance metric between query and the reference vectors in the projection space. Excellent results in this way. When tested on the Omniglot, miniImageNet tieredImageNet datasets, we obtain state of the art classification under various few-shot scenarios.