Summary of Measuring Bias Of Web-filtered Text Datasets and Bias Propagation Through Training, by Youssef Mansour and Reinhard Heckel
Measuring Bias of Web-filtered Text Datasets and Bias Propagation Through Training
by Youssef Mansour, Reinhard Heckel
First submitted to arxiv on: 3 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 investigates biases in pretraining datasets for large language models (LLMs) through dataset classification experiments. It analyzes popular open-source datasets for LLMs, including C4, RefinedWeb, DolmaCC, RedPajama-V2, FineWeb, and DCLM-Baseline, finding that neural networks can surprisingly well classify which dataset a text sequence belongs to. This indicates small differences in filtering and processing pipelines induce distinct fingerprints in formatting, vocabulary, and content distributions. The biases remain even when the text is rewritten with LLMs and propagate through training, enabling estimation of pretraining mixture proportions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how language models are trained using different datasets. It finds that these datasets have tiny differences that make it easy for computers to tell which dataset a piece of text comes from. This means the biases in the datasets can be used to figure out what mix of data sources was used to train the model. |
Keywords
» Artificial intelligence » Classification » Pretraining
