Exploring the Similarity of Representations in Model-Agnostic Meta-Learning

TL;DR

This paper investigates why model-agnostic meta-learning (MAML) performs well by analyzing its representations using neuroscience-inspired similarity analysis, revealing that feature reuse is significant but also influenced by the task and training process.

Contribution

It applies representation similarity analysis to MAML, providing new insights into the roles of feature reuse and task influence in its success.

Findings

Representation similarity increases are partly due to the learning task.

Inner gradient steps cause broader changes than meta-training.

Some evidence supports feature reuse as a key factor, but not entirely.

Abstract

In past years model-agnostic meta-learning (MAML) has been one of the most promising approaches in meta-learning. It can be applied to different kinds of problems, e.g., reinforcement learning, but also shows good results on few-shot learning tasks. Besides their tremendous success in these tasks, it has still not been fully revealed yet, why it works so well. Recent work proposes that MAML rather reuses features than rapidly learns. In this paper, we want to inspire a deeper understanding of this question by analyzing MAML's representation. We apply representation similarity analysis (RSA), a well-established method in neuroscience, to the few-shot learning instantiation of MAML. Although some part of our analysis supports their general results that feature reuse is predominant, we also reveal arguments against their conclusion. The similarity-increase of layers closer to the input layers arises from the learning task itself and not from the model. In addition, the representations after inner gradient steps make a broader change to the representation than the changes during meta-training.