Generalized Inner Loop Meta-Learning
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.
Cite
@article{arxiv.1910.01727,
title = {Generalized Inner Loop Meta-Learning},
author = {Edward Grefenstette and Brandon Amos and Denis Yarats and Phu Mon Htut and Artem Molchanov and Franziska Meier and Douwe Kiela and Kyunghyun Cho and Soumith Chintala},
journal= {arXiv preprint arXiv:1910.01727},
year = {2019}
}
Comments
17 pages, 3 figures, 1 algorithm