Model-based reinforcement learning has recently demonstrated significant advances in solving complex problems of sequential decision-making. Updating the model using the case of solving the current task allows the agent to update the model and apply it to improve the efficiency of solving the following similar tasks. This approach also aligns with case-based planning methods, which already have mechanisms for retrieving and reusing precedents. In this work, we propose a meta-learned case retrieval mechanism that provides case-based samples for the agent to accelerate the learning process. We have tested the performance of the proposed approach on the well-known MuJoCo dataset and have shown results at the level of methods using pre-generated expert data.
Электронная библиотека Ассоциации вычислительной техники: https://dl.acm.org/doi/abs/10.1007/978-3-030-93758-4_35
Zholus A., Panov A. I. Case-based Task Generalization in Model-based Reinforcement Learning // Artificial General Intelligence. AGI 2021. Lecture Notes in Computer Science / ed. Goertzel B., Iklé M., Potapov A. Springer International Publishing, 2022. Vol. 13154. P. 344–354.