Sharing Knowledge in Multi-Task Deep Reinforcement Learning

TL;DR

This paper investigates how sharing representations among multiple tasks in deep reinforcement learning can improve generalization and efficiency, supported by theoretical analysis and empirical results on benchmark tasks.

Contribution

It provides theoretical guarantees for multi-task representation sharing and introduces multi-task extensions of RL algorithms with empirical validation.

Findings

Shared representations improve sample efficiency and performance.

Theoretical conditions for effective sharing are established.

Multi-task algorithms outperform single-task counterparts on benchmarks.

Abstract

We study the benefit of sharing representations among tasks to enable the effective use of deep neural networks in Multi-Task Reinforcement Learning. We leverage the assumption that learning from different tasks, sharing common properties, is helpful to generalize the knowledge of them resulting in a more effective feature extraction compared to learning a single task. Intuitively, the resulting set of features offers performance benefits when used by Reinforcement Learning algorithms. We prove this by providing theoretical guarantees that highlight the conditions for which is convenient to share representations among tasks, extending the well-known finite-time bounds of Approximate Value-Iteration to the multi-task setting. In addition, we complement our analysis by proposing multi-task extensions of three Reinforcement Learning algorithms that we empirically evaluate on widely used Reinforcement Learning benchmarks showing significant improvements over the single-task counterparts in terms of sample efficiency and performance.