Summary of Tso: Self-training with Scaled Preference Optimization, by Kaihui Chen et al.
TSO: Self-Training with Scaled Preference Optimization
by Kaihui Chen, Hao Yi, Qingyang Li, Tianyu Qi, Yulan Hu, Fuzheng Zhang, Yong Liu
First submitted to arxiv on: 31 Aug 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 The proposed framework, TSO (Self-Training with Scaled Preference Optimization), addresses the challenges of large language models’ conformity to human preferences. By introducing a novel approach that conducts self-training preference learning without requiring additional reward models, TSO enhances response diversity through model matrices and incorporates human preference responses. The framework also corrects model preference errors using human and AI feedback, and adopts iterative and dual clip reward strategies to update the reference model and its responses. Experimental results demonstrate that TSO outperforms existing methods on various alignment evaluation benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research proposes a new way to help large language models better match what humans like or dislike. The approach, called TSO, does this by training the model to learn from its own responses, rather than relying on additional reward models. This makes it easier to collect high-quality preference data and improve the model’s performance. |
Keywords
» Artificial intelligence » Alignment » Optimization » Self training
