Summary of Razor: Sharpening Knowledge by Cutting Bias with Unsupervised Text Rewriting, By Shuo Yang et al.
RAZOR: Sharpening Knowledge by Cutting Bias with Unsupervised Text Rewriting
by Shuo Yang, Bardh Prenkaj, Gjergji Kasneci
First submitted to arxiv on: 10 Dec 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Machine Learning (cs.LG)
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 paper proposes RAZOR (Rewriting And Zero-bias Optimization Refinement), an unsupervised debiasing approach based on text rewriting to mitigate shortcuts in pre-trained language models. The method leverages large language models (LLMs) to iteratively rewrite biased text segments with heuristically selected alternatives, aligning surface-level features with diverse label distributions and promoting genuine linguistic patterns. Compared to SoTA models, RAZOR improves F1 scores by 3.5% on the FEVER dataset and 6.5% on MNLI and SNLI datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper talks about a way to make language models more fair. They have a problem where some words are connected with certain meanings because of how data was collected, not because it’s actually true. So they made an algorithm that helps fix this by changing the text to get rid of these biases. This makes the model better at understanding real language and reduces unfairness. |
Keywords
» Artificial intelligence » Optimization » Unsupervised
