Offline Q-Learning on Diverse Multi-Task Data Both Scales And Generalizes
This addresses the problem of scaling offline RL for broader generalization in AI agents, representing a significant advance rather than an incremental improvement.
The paper tackles the challenge of scaling offline reinforcement learning by showing that with appropriate architectural choices (ResNets, cross-entropy distributional backups, feature normalization), offline Q-learning algorithms achieve strong performance that scales with model capacity, achieving near-human performance on 40 Atari games with up to 80 million parameters and extrapolating beyond dataset performance even with suboptimal data.
The potential of offline reinforcement learning (RL) is that high-capacity models trained on large, heterogeneous datasets can lead to agents that generalize broadly, analogously to similar advances in vision and NLP. However, recent works argue that offline RL methods encounter unique challenges to scaling up model capacity. Drawing on the learnings from these works, we re-examine previous design choices and find that with appropriate choices: ResNets, cross-entropy based distributional backups, and feature normalization, offline Q-learning algorithms exhibit strong performance that scales with model capacity. Using multi-task Atari as a testbed for scaling and generalization, we train a single policy on 40 games with near-human performance using up-to 80 million parameter networks, finding that model performance scales favorably with capacity. In contrast to prior work, we extrapolate beyond dataset performance even when trained entirely on a large (400M transitions) but highly suboptimal dataset (51% human-level performance). Compared to return-conditioned supervised approaches, offline Q-learning scales similarly with model capacity and has better performance, especially when the dataset is suboptimal. Finally, we show that offline Q-learning with a diverse dataset is sufficient to learn powerful representations that facilitate rapid transfer to novel games and fast online learning on new variations of a training game, improving over existing state-of-the-art representation learning approaches.