Revisiting Unsupervised Meta-Learning via the Characteristics of Few-Shot Tasks

TL;DR

This paper proposes an unsupervised meta-learning approach for few-shot image classification that leverages task sampling strategies and similarity measures, achieving competitive results without labeled base classes.

Contribution

It introduces a novel unsupervised meta-learning framework that removes the need for labeled base classes and enhances performance through task sampling, similarity normalization, and auxiliary embedding transformations.

Findings

Outperforms previous unsupervised meta-learning methods.

Achieves comparable or better results than supervised approaches.

Effective use of synthesized confusing instances and task-specific transformations.

Abstract

Meta-learning has become a practical approach towards few-shot image classification, where "a strategy to learn a classifier" is meta-learned on labeled base classes and can be applied to tasks with novel classes. We remove the requirement of base class labels and learn generalizable embeddings via Unsupervised Meta-Learning (UML). Specifically, episodes of tasks are constructed with data augmentations from unlabeled base classes during meta-training, and we apply embedding-based classifiers to novel tasks with labeled few-shot examples during meta-test. We observe two elements play important roles in UML, i.e., the way to sample tasks and measure similarities between instances. Thus we obtain a strong baseline with two simple modifications -- a sufficient sampling strategy constructing multiple tasks per episode efficiently together with a semi-normalized similarity. We then take advantage of the characteristics of tasks from two directions to get further improvements. First, synthesized confusing instances are incorporated to help extract more discriminative embeddings. Second, we utilize an additional task-specific embedding transformation as an auxiliary component during meta-training to promote the generalization ability of the pre-adapted embeddings. Experiments on few-shot learning benchmarks verify that our approaches outperform previous UML methods and achieve comparable or even better performance than its supervised variants.