Interpretable Dynamics Models for Data-Efficient Reinforcement Learning
This work addresses data-efficiency challenges in reinforcement learning for domains requiring interpretability, though it appears incremental as it builds on existing Bayesian and model-based methods.
The paper tackled the problem of data-inefficient reinforcement learning by proposing a Bayesian model-based approach that incorporates expert knowledge into structured transition models, resulting in improved data-efficiency and human-interpretable insights on a heteroskedastic and bimodal benchmark, with comparisons to NFQ.
In this paper, we present a Bayesian view on model-based reinforcement learning. We use expert knowledge to impose structure on the transition model and present an efficient learning scheme based on variational inference. This scheme is applied to a heteroskedastic and bimodal benchmark problem on which we compare our results to NFQ and show how our approach yields human-interpretable insight about the underlying dynamics while also increasing data-efficiency.