Summary of Adaptive Segment-level Reward: Bridging the Gap Between Action and Reward Space in Alignment, by Yanshi Li et al.
Adaptive Segment-level Reward: Bridging the Gap Between Action and Reward Space in Alignment
by Yanshi Li, Shaopan Xiong, Gengru Chen, Xiaoyang Li, Yijia Luo, Xingyuan Bu, Yingshui Tan, Wenbo Su, Bo Zheng
First submitted to arxiv on: 23 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
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 an innovative reinforcement learning (RL) method to improve credit assignment in aligning large language models with human preferences. The traditional RL methods optimize sequence rewards, but this can lead to suboptimal learning. To address this, the authors introduce the “Adaptive Segment-wise Reward” approach, which uses semantic meaning to adaptively segment tokens. This method is integrated into various training methods and improves success rates on adversarial samples by 10% and evaluation benchmarks like MMLU, GSM8K, HumanEval, etc. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about using a new way of teaching computers to understand human language better. Right now, computer models are very good at generating text, but they don’t always get it right. The authors want to make these models better by giving them rewards when they do something good. They came up with a new way of doing this that uses the meaning of words instead of just punctuation marks. This helps the model focus on the most important parts of language and makes it better at understanding human language. |
Keywords
* Artificial intelligence * Reinforcement learning
