ES-MAML: Simple Hessian-Free Meta Learning
This addresses a bottleneck in meta-learning for researchers and practitioners by providing a simpler, more efficient alternative to gradient-based MAML methods, though it is incremental as it builds on existing ES and MAML concepts.
The paper tackles the difficulty of estimating second derivatives in Model-Agnostic Meta-Learning (MAML) by introducing ES-MAML, a framework based on Evolution Strategies (ES) that avoids this issue and is simple to implement. The result shows that ES-MAML is competitive with existing methods and often achieves better adaptation with fewer queries.
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.