Generalized Inner Loop Meta-Learning

TL;DR

This paper formalizes a common pattern in meta-learning called GIMLI, providing a general algorithm and library to facilitate future research and practical applications in deep learning and reinforcement learning.

Contribution

It introduces GIMLI, a formalization and general algorithm for nested optimization in meta-learning, along with a supportive library called higher.

Findings

Demonstrates the practical utility of GIMLI through experiments

Provides ablation studies to analyze the framework's components

Offers a software library to implement GIMLI-based meta-learning approaches

Abstract

Many (but not all) approaches self-qualifying as "meta-learning" in deep learning and reinforcement learning fit a common pattern of approximating the solution to a nested optimization problem. In this paper, we give a formalization of this shared pattern, which we call GIMLI, prove its general requirements, and derive a general-purpose algorithm for implementing similar approaches. Based on this analysis and algorithm, we describe a library of our design, higher, which we share with the community to assist and enable future research into these kinds of meta-learning approaches. We end the paper by showcasing the practical applications of this framework and library through illustrative experiments and ablation studies which they facilitate.