Summary of Measuring Memorization in Language Models Via Probabilistic Extraction, by Jamie Hayes et al.
Measuring memorization in language models via probabilistic extraction
by Jamie Hayes, Marika Swanberg, Harsh Chaudhari, Itay Yona, Ilia Shumailov, Milad Nasr, Christopher A. Choquette-Choo, Katherine Lee, A. Feder Cooper
First submitted to arxiv on: 25 Oct 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 This paper investigates the reliability of measuring memorization in large language models (LLMs) using a method called Discoverable Extraction. The authors show that this approach has limitations when dealing with non-deterministic sampling schemes, which are more common in real-world scenarios. They introduce Probabilistic Discoverable Extraction, a revised measure that considers multiple queries to quantify the probability of extracting sensitive information. This approach provides a more nuanced understanding of extraction risk and is evaluated across different models, sampling schemes, and training-data repetitions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can remember what they learned during training, which raises concerns about accidentally sharing private information. One way to check if this happens is by testing how well the model recalls specific parts of its training data. The current method for doing this has some limitations. The authors propose a new approach that considers multiple possible outcomes when checking if the model remembers something. This helps give a more accurate idea of the risk of sharing sensitive information. |
Keywords
» Artificial intelligence » Probability
