What is Going on Inside Recurrent Meta Reinforcement Learning Agents?

TL;DR

This paper investigates the internal mechanisms of recurrent meta reinforcement learning agents, proposing they function as belief state learners in partially observable environments, which enhances understanding of their behavior and limitations.

Contribution

The study reformulates meta-RL using POMDPs and demonstrates that RNN activity dynamics act as belief states, providing new insights into how these agents learn and operate.

Findings

RNN activity functions as belief states in meta-RL agents.

Meta-RL agents can be viewed as learning to act in POMDPs.

Insights into failure modes and model-based results of meta-RL agents.

Abstract

Recurrent meta reinforcement learning (meta-RL) agents are agents that employ a recurrent neural network (RNN) for the purpose of "learning a learning algorithm". After being trained on a pre-specified task distribution, the learned weights of the agent's RNN are said to implement an efficient learning algorithm through their activity dynamics, which allows the agent to quickly solve new tasks sampled from the same distribution. However, due to the black-box nature of these agents, the way in which they work is not yet fully understood. In this study, we shed light on the internal working mechanisms of these agents by reformulating the meta-RL problem using the Partially Observable Markov Decision Process (POMDP) framework. We hypothesize that the learned activity dynamics is acting as belief states for such agents. Several illustrative experiments suggest that this hypothesis is true, and that recurrent meta-RL agents can be viewed as agents that learn to act optimally in partially observable environments consisting of multiple related tasks. This view helps in understanding their failure cases and some interesting model-based results reported in the literature.