Summary of Certification For Differentially Private Prediction in Gradient-based Training, by Matthew Wicker et al.
Certification for Differentially Private Prediction in Gradient-Based Training
by Matthew Wicker, Philip Sosnin, Igor Shilov, Adrianna Janik, Mark N. Müller, Yves-Alexandre de Montjoye, Adrian Weller, Calvin Tsay
First submitted to arxiv on: 19 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 research paper proposes a novel approach to provide meaningful differential privacy guarantees for machine learning models at prediction time. The private prediction setting privatizes model outputs to ensure formal guarantees. However, existing algorithms rely on global sensitivity, which can result in excessive noise addition. To address this challenge, the authors introduce a practical framework based on convex relaxation and bound propagation to compute provable upper-bounds for local and smooth sensitivities. This enables reduced noise addition or improved privacy accounting in private prediction settings. The approach is validated across datasets from various domains (financial services, medical image classification, natural language processing) and models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps ensure that machine learning models keep people’s information safe. Currently, it’s hard to guarantee privacy because existing methods add too much noise to the predictions. The researchers came up with a new way to calculate how much noise is needed so we can add less or even improve our accounting of privacy. They tested their approach on various data sets and found that it works well. |
Keywords
» Artificial intelligence » Image classification » Machine learning » Natural language processing
