Summary of Self-training Meets Consistency: Improving Llms’ Reasoning with Consistency-driven Rationale Evaluation, by Jaehyeok Lee et al.
Self-Training Meets Consistency: Improving LLMs’ Reasoning with Consistency-Driven Rationale Evaluation
by Jaehyeok Lee, Keisuke Sakaguchi, JinYeong Bak
First submitted to arxiv on: 10 Nov 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 research paper proposes a novel approach to training large language models (LLMs) called CREST, which aims to improve their reasoning abilities by evaluating the quality of their self-generated rationales. The authors argue that previous methods for labeling rationals as correct or not may lead to flawed learning patterns if they rely on a single measure. To overcome this issue, CREST introduces two methods: filtering out rationals that often result in incorrect answers and preference learning based on mixed preferences from rationale evaluation results. Experimental results on three question-answering datasets show that CREST improves the logical robustness and correctness of rationales, as well as reasoning abilities compared to previous approaches. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can get better at giving good reasons for their answers when they’re trained on their own ideas. Right now, researchers are trying different ways to teach these models what makes a good reason. This paper introduces a new approach called CREST that looks at how well the model’s reasons hold up under more questions. It shows that this method helps the models come up with better, more logical answers. |
Keywords
» Artificial intelligence » Question answering
