Summary of Exposing Privacy Gaps: Membership Inference Attack on Preference Data For Llm Alignment, by Qizhang Feng et al.
Exposing Privacy Gaps: Membership Inference Attack on Preference Data for LLM Alignment
by Qizhang Feng, Siva Rajesh Kasa, Hyokun Yun, Choon Hui Teo, Sravan Babu Bodapati
First submitted to arxiv on: 8 Jul 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- 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 A novel approach to refining Large Language Models (LLMs) is proposed, which utilizes human preference data through methods like Proximal Policy Optimization (PPO) and Direct Preference Optimization (DPO). However, the use of such data raises privacy concerns. This paper investigates the vulnerability of LLMs aligned using human preference datasets to membership inference attacks (MIAs), introducing a novel reference-based attack framework called PREMIA. The study finds that DPO models are more susceptible to MIAs compared to PPO models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Language Models can generate text, but they need to be trained to follow human standards. Some methods, like Proximal Policy Optimization and Direct Preference Optimization, help refine these models using human preferences. However, this raises concerns about privacy. This study looks at how vulnerable these refined models are to attacks that try to figure out if a model has seen specific data before. The researchers introduce a new way to analyze preference data called PREMIA and find that some methods are more vulnerable than others. |
Keywords
» Artificial intelligence » Inference » Optimization
