Minimal Value-Equivalent Partial Models for Scalable and Robust Planning in Lifelong Reinforcement Learning
This addresses the challenge of scalable and robust planning for lifelong reinforcement learning agents, representing an incremental improvement by refining model-based approaches.
The paper tackles the problem of learning comprehensive environment models in lifelong reinforcement learning, which can be inefficient, by proposing minimal value-equivalent partial models that focus only on relevant aspects. The result shows these models offer scalability advantages and robustness to distribution shifts and model errors, as supported by theoretical and empirical evidence.
Learning models of the environment from pure interaction is often considered an essential component of building lifelong reinforcement learning agents. However, the common practice in model-based reinforcement learning is to learn models that model every aspect of the agent's environment, regardless of whether they are important in coming up with optimal decisions or not. In this paper, we argue that such models are not particularly well-suited for performing scalable and robust planning in lifelong reinforcement learning scenarios and we propose new kinds of models that only model the relevant aspects of the environment, which we call "minimal value-equivalent partial models". After providing a formal definition for these models, we provide theoretical results demonstrating the scalability advantages of performing planning with such models and then perform experiments to empirically illustrate our theoretical results. Then, we provide some useful heuristics on how to learn these kinds of models with deep learning architectures and empirically demonstrate that models learned in such a way can allow for performing planning that is robust to distribution shifts and compounding model errors. Overall, both our theoretical and empirical results suggest that minimal value-equivalent partial models can provide significant benefits to performing scalable and robust planning in lifelong reinforcement learning scenarios.