Boosting Supervision with Self-Supervision for Few-shot Learning

TL;DR

This paper introduces a method that combines self-supervised learning with supervised training to enhance the transferability and generalization of deep representations, especially in few-shot learning scenarios, without needing external data.

Contribution

The paper proposes a novel approach that integrates self-supervised tasks as auxiliary losses to improve deep representation transferability on small datasets and in few-shot learning.

Findings

Self-supervised losses reduce error rates by 5-25% on few-shot benchmarks.

Benefits of self-supervision increase with task difficulty.

Method improves generalization on standard classification tasks without external data.

Abstract

We present a technique to improve the transferability of deep representations learned on small labeled datasets by introducing self-supervised tasks as auxiliary loss functions. While recent approaches for self-supervised learning have shown the benefits of training on large unlabeled datasets, we find improvements in generalization even on small datasets and when combined with strong supervision. Learning representations with self-supervised losses reduces the relative error rate of a state-of-the-art meta-learner by 5-25% on several few-shot learning benchmarks, as well as off-the-shelf deep networks on standard classification tasks when training from scratch. We find the benefits of self-supervision increase with the difficulty of the task. Our approach utilizes the images within the dataset to construct self-supervised losses and hence is an effective way of learning transferable representations without relying on any external training data.