ES-MAML: Simple Hessian-Free Meta Learning

TL;DR

ES-MAML introduces a simple, Hessian-free meta learning framework using Evolution Strategies, avoiding second derivative estimation issues and improving adaptation efficiency over existing gradient-based methods.

Contribution

The paper presents ES-MAML, a novel meta learning approach based on Evolution Strategies that simplifies implementation and extends applicability to nonsmooth adaptation operators.

Findings

ES-MAML is competitive with existing meta learning methods.

It often achieves better adaptation with fewer queries.

The method avoids complex second derivative estimation.

Abstract

We introduce ES-MAML, a new framework for solving the model agnostic meta learning (MAML) problem based on Evolution Strategies (ES). Existing algorithms for MAML are based on policy gradients, and incur significant difficulties when attempting to estimate second derivatives using backpropagation on stochastic policies. We show how ES can be applied to MAML to obtain an algorithm which avoids the problem of estimating second derivatives, and is also conceptually simple and easy to implement. Moreover, ES-MAML can handle new types of nonsmooth adaptation operators, and other techniques for improving performance and estimation of ES methods become applicable. We show empirically that ES-MAML is competitive with existing methods and often yields better adaptation with fewer queries.