Summary of Data Contamination Calibration For Black-box Llms, by Wentao Ye et al.
Data Contamination Calibration for Black-box LLMs
by Wentao Ye, Jiaqi Hu, Liyao Li, Haobo Wang, Gang Chen, Junbo Zhao
First submitted to arxiv on: 20 May 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 proposed Polarized Augment Calibration (PAC) method tackles the issue of unchecked ultra-large-scale training data by detecting contaminated data and minimizing its effect. This holistic approach extends the Membership Inference Attack (MIA) to form a global target for detecting invisible training data, making it plug-and-play with most current LLMs. PAC outperforms existing methods by at least 4.5% on four dataset formats using over 10 base LLMs. The authors also demonstrate the real-world impact of contamination and related issues. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to detect contaminated training data for Large Language Models (LLMs). This is important because huge amounts of data are being used to train these models, but some of this data might be fake or biased. The proposed method, called PAC, can be used with many different LLMs and has been shown to work better than other methods in detecting contaminated data. |
Keywords
» Artificial intelligence » Inference
