Summary of Learning From Imperfect Human Feedback: a Tale From Corruption-robust Dueling, by Yuwei Cheng et al.
Learning from Imperfect Human Feedback: a Tale from Corruption-Robust Dueling
by Yuwei Cheng, Fan Yao, Xuefeng Liu, Haifeng Xu
First submitted to arxiv on: 18 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Machine Learning (stat.ML)
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 addresses the issue of learning from imperfect human feedback, where humans may not always provide accurate or consistent feedback. The authors model this problem as a continuous-action dueling bandit with decaying corruption over time, which means that humans learn to improve their feedback over time. The study shows that even when the rate of imperfection is known, there is still a regret lower bound for Learning from Imperfect Human Feedback (LIHF). To address this challenge, the authors develop the Robustified Stochastic Mirror Descent for Imperfect Dueling (RoSMID) algorithm, which achieves nearly optimal regret. The paper’s analysis is based on a novel framework that can be applied to other gradient-based dueling bandit algorithms. Experimental results validate the theoretical findings. |
Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how machines learn from people who give them feedback. Sometimes humans don’t give the best feedback because they’re not perfect either. The authors want to know if it’s possible for a machine to still learn from this imperfect feedback. They create a special model that shows even when humans are trying their best, there’s still room for improvement. To help with this problem, the researchers develop a new algorithm that can handle imperfect feedback. This algorithm is tested and shown to work well in real-world scenarios. |