Summary of Sample-efficient Preference-based Reinforcement Learning with Dynamics Aware Rewards, by Katherine Metcalf et al.
Sample-Efficient Preference-based Reinforcement Learning with Dynamics Aware Rewards
by Katherine Metcalf, Miguel Sarabia, Natalie Mackraz, Barry-John Theobald
First submitted to arxiv on: 28 Feb 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Machine Learning (cs.LG)
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 In this paper, the authors propose a novel approach to preference-based reinforcement learning (PbRL) that leverages dynamics-aware reward functions. By learning a state-action representation using self-supervised temporal consistency tasks and bootstrapping the reward function from it, they demonstrate significant improvements in sample efficiency and policy performance. Specifically, their method achieves comparable results to existing approaches with 5 times fewer preference labels. This advance has implications for robotic control and other applications where human feedback is limited. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This study shows how robots can learn to behave according to human preferences by using a special type of reward function. The researchers did this by combining two things: learning about the robot’s actions and states, and then using that information to create a better reward function. This new approach made the robot’s training much faster and more effective. With only 50 preference labels, it achieved the same results as other methods that needed 500 labels. |
Keywords
* Artificial intelligence * Bootstrapping * Reinforcement learning * Self supervised
