Learning to Modulate pre-trained Models in RL

TL;DR

This paper investigates catastrophic forgetting in RL when fine-tuning pre-trained models on new tasks, and proposes a novel modulation method that retains pre-trained skills while adapting to new tasks, achieving state-of-the-art results.

Contribution

The paper introduces Learning-to-Modulate (L2M), a new method that prevents forgetting in RL fine-tuning by modulating a frozen pre-trained model's information flow.

Findings

L2M outperforms existing fine-tuning methods on Continual-World.

Most fine-tuning approaches cause significant performance degradation on pre-training tasks.

L2M retains pre-trained skills while achieving high performance on new tasks.

Abstract

Reinforcement Learning (RL) has been successful in various domains like robotics, game playing, and simulation. While RL agents have shown impressive capabilities in their specific tasks, they insufficiently adapt to new tasks. In supervised learning, this adaptation problem is addressed by large-scale pre-training followed by fine-tuning to new down-stream tasks. Recently, pre-training on multiple tasks has been gaining traction in RL. However, fine-tuning a pre-trained model often suffers from catastrophic forgetting. That is, the performance on the pre-training tasks deteriorates when fine-tuning on new tasks. To investigate the catastrophic forgetting phenomenon, we first jointly pre-train a model on datasets from two benchmark suites, namely Meta-World and DMControl. Then, we evaluate and compare a variety of fine-tuning methods prevalent in natural language processing, both in terms of performance on new tasks, and how well performance on pre-training tasks is retained. Our study shows that with most fine-tuning approaches, the performance on pre-training tasks deteriorates significantly. Therefore, we propose a novel method, Learning-to-Modulate (L2M), that avoids the degradation of learned skills by modulating the information flow of the frozen pre-trained model via a learnable modulation pool. Our method achieves state-of-the-art performance on the Continual-World benchmark, while retaining performance on the pre-training tasks. Finally, to aid future research in this area, we release a dataset encompassing 50 Meta-World and 16 DMControl tasks.