Summary of Free Record-level Privacy Risk Evaluation Through Artifact-based Methods, by Joseph Pollock et al.
Free Record-Level Privacy Risk Evaluation Through Artifact-Based Methods
by Joseph Pollock, Igor Shilov, Euodia Dodd, Yves-Alexandre de Montjoye
First submitted to arxiv on: 8 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR)
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 Machine learning models can be vulnerable to membership inference attacks (MIAs), which are used to assess privacy risks. State-of-the-art methods require training many shadow models, making them impractical for large models or iterative use during development. The proposed Loss Trace Interquantile Range (LT-IQR) method analyzes per-sample loss trajectories collected during model training to identify high-risk samples without requiring additional model training. LT-IQR achieves 92% precision@k=1% in identifying vulnerable samples across datasets and model architectures, outperforming traditional vulnerability metrics and lightweight MIAs. This enables model developers to efficiently evaluate privacy risks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Machine learning models can be vulnerable to attacks that try to figure out if a specific sample was used during training. To do this, attackers train many fake models and compare how well they perform on the sample in question. This is time-consuming and impractical for large models or when testing different versions of a model. A new way to identify these vulnerable samples without creating fake models has been developed. It works by looking at how each sample performed during training and finds the ones that were most affected. This method, called LT-IQR, was tested on several datasets and found to be very accurate. |
Keywords
* Artificial intelligence * Inference * Machine learning * Precision
