Generalization of Reinforcement Learners with Working and Episodic Memory

TL;DR

This paper develops a methodology to evaluate how different memory systems in reinforcement learning agents generalize to holdout data, using diverse tasks and ablation studies to analyze performance.

Contribution

It introduces a comprehensive testing framework for memory in reinforcement learning, including diverse tasks and ablation analysis of combined memory systems.

Findings

Memory systems improve generalization on specific tasks

Combined memory architectures outperform single systems

Evaluation methodology reveals strengths and limitations of memory types

Abstract

Memory is an important aspect of intelligence and plays a role in many deep reinforcement learning models. However, little progress has been made in understanding when specific memory systems help more than others and how well they generalize. The field also has yet to see a prevalent consistent and rigorous approach for evaluating agent performance on holdout data. In this paper, we aim to develop a comprehensive methodology to test different kinds of memory in an agent and assess how well the agent can apply what it learns in training to a holdout set that differs from the training set along dimensions that we suggest are relevant for evaluating memory-specific generalization. To that end, we first construct a diverse set of memory tasks that allow us to evaluate test-time generalization across multiple dimensions. Second, we develop and perform multiple ablations on an agent architecture that combines multiple memory systems, observe its baseline models, and investigate its performance against the task suite.