A Simple Yet Effective Strategy to Robustify the Meta Learning Paradigm
This work addresses robustness issues in meta learning for risk-sensitive scenarios, representing an incremental improvement over existing methods.
The paper tackles the problem of catastrophic worst-case fast adaptation in meta learning by optimizing pipelines from a distributionally robust perspective, using a two-stage strategy to control worst-case risks at a probabilistic level, resulting in improved robustness to task distributions and reduced conditional expectation of worst-case adaptation risk.
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.