arXiv · Machine Learning Theory· Mintae Kim, Koushil Sreenath·· 3 小时前AI 评分22
In-Distribution Imagination for Model-Based Offline Reinforcement Learning
In-Distribution Imagination for Model-Based Offline Reinforcement Learning
AI 导读
提出 in-distribution imagination（IDI）框架，通过在 learned representation space 中估计 trajectory support，自适应截断偏离 offline trajectory manifold 的 rollouts。
来源：arXiv · Machine Learning Theory · arxiv.org