Is Bayesian Model-Agnostic Meta Learning Better than Model-Agnostic Meta Learning, Provably?
This provides theoretical justification for Bayesian MAML's empirical advantages, addressing a gap in meta-learning theory for researchers in the field.
The paper tackles the theoretical understanding of Bayesian Model-Agnostic Meta Learning (MAML) compared to standard MAML, establishing that Bayesian MAML has provably lower meta test risks in meta linear regression under specific cases, with experimental verification.
Meta learning aims at learning a model that can quickly adapt to unseen tasks. Widely used meta learning methods include model agnostic meta learning (MAML), implicit MAML, Bayesian MAML. Thanks to its ability of modeling uncertainty, Bayesian MAML often has advantageous empirical performance. However, the theoretical understanding of Bayesian MAML is still limited, especially on questions such as if and when Bayesian MAML has provably better performance than MAML. In this paper, we aim to provide theoretical justifications for Bayesian MAML's advantageous performance by comparing the meta test risks of MAML and Bayesian MAML. In the meta linear regression, under both the distribution agnostic and linear centroid cases, we have established that Bayesian MAML indeed has provably lower meta test risks than MAML. We verify our theoretical results through experiments.