Stabilizing Transformers for Reinforcement Learning

TL;DR

This paper introduces Gated Transformer-XL, an improved transformer architecture tailored for reinforcement learning, which enhances stability, learning speed, and performance in memory-intensive and partially observable tasks.

Contribution

The authors propose architectural modifications to the Transformer-XL that significantly improve its stability and efficiency in reinforcement learning settings, outperforming LSTMs and previous architectures.

Findings

GTrXL surpasses LSTMs on memory environments.

Achieves state-of-the-art on DMLab-30 benchmark.

Offers a more expressive alternative to LSTMs for RL.

Abstract

Owing to their ability to both effectively integrate information over long time horizons and scale to massive amounts of data, self-attention architectures have recently shown breakthrough success in natural language processing (NLP), achieving state-of-the-art results in domains such as language modeling and machine translation. Harnessing the transformer's ability to process long time horizons of information could provide a similar performance boost in partially observable reinforcement learning (RL) domains, but the large-scale transformers used in NLP have yet to be successfully applied to the RL setting. In this work we demonstrate that the standard transformer architecture is difficult to optimize, which was previously observed in the supervised learning setting but becomes especially pronounced with RL objectives. We propose architectural modifications that substantially improve the stability and learning speed of the original Transformer and XL variant. The proposed architecture, the Gated Transformer-XL (GTrXL), surpasses LSTMs on challenging memory environments and achieves state-of-the-art results on the multi-task DMLab-30 benchmark suite, exceeding the performance of an external memory architecture. We show that the GTrXL, trained using the same losses, has stability and performance that consistently matches or exceeds a competitive LSTM baseline, including on more reactive tasks where memory is less critical. GTrXL offers an easy-to-train, simple-to-implement but substantially more expressive architectural alternative to the standard multi-layer LSTM ubiquitously used for RL agents in partially observable environments.