Summary of Scaling Up Membership Inference: When and How Attacks Succeed on Large Language Models, by Haritz Puerto et al.
Scaling Up Membership Inference: When and How Attacks Succeed on Large Language Models
by Haritz Puerto, Martin Gubri, Sangdoo Yun, Seong Joon Oh
First submitted to arxiv on: 31 Oct 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 This paper challenges recent claims that membership inference attacks (MIA) no longer work on large language models (LLMs). The authors argue that MIA can still be effective, but only when testing multiple documents. They construct new benchmarks to measure MIA performance at different scales, from individual sentences to document collections. To validate current approaches, they adapt a recent dataset inference (DI) method for binary membership detection and fine-tune it for MIA on pre-trained and fine-tuned LLMs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Membership inference attacks can detect whether a data sample is part of the training set for a large language model. Some people are worried about these models being trained with copyrighted materials, so they need ways to spot this usage. But recent research said that current methods don’t work on large language models. The authors think that’s not true – MIA can still work, but only if you test multiple documents at once. They created new tests to see how well different methods do. |
Keywords
» Artificial intelligence » Inference » Large language model
