Summary of Robust Reinforcement Learning From Corrupted Human Feedback, by Alexander Bukharin et al.
Robust Reinforcement Learning from Corrupted Human Feedback
by Alexander Bukharin, Ilgee Hong, Haoming Jiang, Zichong Li, Qingru Zhang, Zixuan Zhang, Tuo Zhao
First submitted to arxiv on: 21 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 This paper proposes a robust reinforcement learning from human feedback approach, called R^3M, which addresses the challenge of incorrect or inconsistent preference labels in AI systems. The method models potential corrupted labels as sparse outliers and formulates the reward learning problem as an _1-regularized maximum likelihood estimation problem. An efficient alternating optimization algorithm is developed to solve this problem, with only a negligible increase in computational overhead compared to standard RLHF. Theoretical guarantees are provided for the approach’s consistency in learning the underlying reward and identifying outliers. Experiments on robotic control and natural language generation demonstrate R^3M’s improved robustness against various perturbations of preference data. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps create AI systems that work well with human feedback. Sometimes, people who help train these systems might make mistakes or be inconsistent. The researchers came up with a way to handle this problem called R^3M. It looks at the bad data and treats it like an outlier. They then use a special math problem to learn what the AI should do next. This approach is good because it works well even when the feedback is wrong. Tests showed that R^3M can help make better robots and language systems. |
Keywords
» Artificial intelligence » Likelihood » Optimization » Reinforcement learning from human feedback » Rlhf
