Summary of Lease: Offline Preference-based Reinforcement Learning with High Sample Efficiency, by Xiao-yin Liu et al.
LEASE: Offline Preference-based Reinforcement Learning with High Sample Efficiency
by Xiao-Yin Liu, Guotao Li, Xiao-Hu Zhou, Zeng-Guang Hou
First submitted to arxiv on: 30 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed LEASE algorithm combines a learned transition model to generate unlabeled preference data and an uncertainty-aware mechanism to ensure the accuracy of reward models. By leveraging this approach, the algorithm achieves high sample efficiency in offline preference-based reinforcement learning (PbRL) while overcoming challenges in designing rewards and acquiring human feedback. The LEASE algorithm also provides a theoretical guarantee for policy improvement and generalization bounds for reward models. Experimental results demonstrate comparable performance to baselines under fewer preference data without online interaction. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Offline reinforcement learning can be challenging because it’s hard to design good rewards and get enough data. Researchers propose an algorithm called LEASE that helps overcome these challenges. It uses a learned model to generate more data, which is then used to improve the reward system. The algorithm also checks how sure it is about its predictions and only uses the most confident ones. This makes the algorithm better at learning from small amounts of data without needing human feedback. The results show that LEASE can perform just as well as other algorithms with less data. |
Keywords
* Artificial intelligence * Generalization * Reinforcement learning
