A Simple Yet Effective Strategy to Robustify the Meta Learning Paradigm

TL;DR

This paper introduces a distributionally robust approach to meta learning, optimizing for worst-case fast adaptation risks to improve robustness across diverse task distributions.

Contribution

It proposes a novel robust meta learning framework that minimizes tail risk, enhancing resilience to task distribution shifts compared to traditional empirical risk minimization methods.

Findings

Improves robustness of meta learning to task distribution variations

Reduces the worst-case fast adaptation risk

Enhances stability in risk-sensitive scenarios

Abstract

Meta learning is a promising paradigm to enable skill transfer across tasks. Most previous methods employ the empirical risk minimization principle in optimization. However, the resulting worst fast adaptation to a subset of tasks can be catastrophic in risk-sensitive scenarios. To robustify fast adaptation, this paper optimizes meta learning pipelines from a distributionally robust perspective and meta trains models with the measure of expected tail risk. We take the two-stage strategy as heuristics to solve the robust meta learning problem, controlling the worst fast adaptation cases at a certain probabilistic level. Experimental results show that our simple method can improve the robustness of meta learning to task distributions and reduce the conditional expectation of the worst fast adaptation risk.