Regression Networks for Meta-Learning Few-Shot Classification

This work proposes to meta-learn classification of embedded points by regressing the closest approximation in every class subspace while using the regression error as a distance metric.