Meta-trained agents implement Bayes-optimal agents
Abstract
Memory-based meta-learning is a powerful technique to build agents that adapt fast to any task within a target distribution. A previous theoretical study has argued that this remarkable performance is because the meta-training protocol incentivises agents to behave Bayes-optimally. We empirically investigate this claim on a number of prediction and bandit tasks. Inspired by ideas from theoretical computer science, we show that meta-learned and Bayes-optimal agents not only behave alike, but they even share a similar computational structure, in the sense that one agent system can approximately simulate the other. Furthermore, we show that Bayes-optimal agents are fixed points of the meta-learning dynamics. Our results suggest that memory-based meta-learning might serve as a general technique for numerically approximating Bayes-optimal agents - that is, even for task distributions for which we currently don't possess tractable models.
Cite
@article{arxiv.2010.11223,
title = {Meta-trained agents implement Bayes-optimal agents},
author = {Vladimir Mikulik and Grégoire Delétang and Tom McGrath and Tim Genewein and Miljan Martic and Shane Legg and Pedro A. Ortega},
journal= {arXiv preprint arXiv:2010.11223},
year = {2020}
}
Comments
Published at 34th Conference on Neural Information Processing Systems (NeurIPS 2020), Vancouver, Canada