Summary of Autonomous Data Selection with Zero-shot Generative Classifiers For Mathematical Texts, by Yifan Zhang et al.
Autonomous Data Selection with Zero-shot Generative Classifiers for Mathematical Texts
by Yifan Zhang, Yifan Luo, Yang Yuan, Andrew Chi-Chih Yao
First submitted to arxiv on: 12 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); 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 presents Autonomous Data Selection (AutoDS), a method that utilizes base language models as zero-shot “generative classifiers” to automatically curate high-quality mathematical texts. Unlike previous approaches, AutoDS relies solely on the model’s logits to determine whether a passage is mathematically informative and educational. By integrating AutoDS into a continual pretraining pipeline, the authors achieve substantial boosts in downstream performance on challenging math benchmarks (MATH, GSM8K, and BBH) using far fewer tokens than previous methods. The approach achieves roughly a twofold improvement in pretraining token efficiency over strong baselines, underscoring the potential of self-directed data selection in enhancing mathematical reasoning. To facilitate future research, the authors release their curated AutoMathText dataset and provide code availability. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper talks about making machines help humans learn math better. It shows a new way to pick good math texts using language models that don’t need training or human labels. This approach is faster and uses less data than before, which helps improve how well the machine learns math. The authors are sharing their work so others can use it too. |
Keywords
* Artificial intelligence * Logits * Pretraining * Token * Zero shot
