Optimal Dynamic Regret by Transformers for Non-Stationary Reinforcement Learning
This addresses the challenge of non-stationary reinforcement learning for AI systems, though it appears incremental as it builds on known transformer capabilities.
The paper tackles the problem of transformers performing reinforcement learning in non-stationary environments, showing they can achieve nearly optimal dynamic regret bounds and match or outperform existing expert algorithms in experiments.
Transformers have demonstrated exceptional performance across a wide range of domains. While their ability to perform reinforcement learning in-context has been established both theoretically and empirically, their behavior in non-stationary environments remains less understood. In this study, we address this gap by showing that transformers can achieve nearly optimal dynamic regret bounds in non-stationary settings. We prove that transformers are capable of approximating strategies used to handle non-stationary environments and can learn the approximator in the in-context learning setup. Our experiments further show that transformers can match or even outperform existing expert algorithms in such environments.