Recasting Gradient-Based Meta-Learning as Hierarchical Bayes

TL;DR

This paper reinterprets the MAML meta-learning algorithm as a hierarchical Bayesian inference method, providing theoretical insights and practical improvements for complex models using scalable gradient-based inference techniques.

Contribution

It reformulates MAML within a hierarchical Bayesian framework, enabling better understanding and enhancements through approximate inference and curvature estimation.

Findings

MAML can be viewed as hierarchical Bayes inference.

The reformulation allows for scalable gradient-based posterior inference.

Proposed improvements enhance MAML's efficiency and applicability.

Abstract

Meta-learning allows an intelligent agent to leverage prior learning episodes as a basis for quickly improving performance on a novel task. Bayesian hierarchical modeling provides a theoretical framework for formalizing meta-learning as inference for a set of parameters that are shared across tasks. Here, we reformulate the model-agnostic meta-learning algorithm (MAML) of Finn et al. (2017) as a method for probabilistic inference in a hierarchical Bayesian model. In contrast to prior methods for meta-learning via hierarchical Bayes, MAML is naturally applicable to complex function approximators through its use of a scalable gradient descent procedure for posterior inference. Furthermore, the identification of MAML as hierarchical Bayes provides a way to understand the algorithm's operation as a meta-learning procedure, as well as an opportunity to make use of computational strategies for efficient inference. We use this opportunity to propose an improvement to the MAML algorithm that makes use of techniques from approximate inference and curvature estimation.