Summary of Perturb-and-project: Differentially Private Similarities and Marginals, by Vincent Cohen-addad et al.
Perturb-and-Project: Differentially Private Similarities and Marginals
by Vincent Cohen-Addad, Tommaso d’Orsi, Alessandro Epasto, Vahab Mirrokni, Peilin Zhong
First submitted to arxiv on: 7 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR); Data Structures and Algorithms (cs.DS)
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 The paper proposes novel efficient algorithms for privately releasing pair-wise cosine similarities and computing k-way marginal queries over n features using the input perturbations framework for differential privacy. The approach adds noise to the input A in the space of admissible datasets S, then projects back to S. The authors design new methods that achieve comparable guarantees as prior work but with improved efficiency, particularly for t-sparse datasets where t is bounded by n^(5/6)/log n. Additionally, the paper provides a theoretical perspective on why fast input perturbation algorithms work well in practice. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper improves our understanding of differential privacy by creating new ways to privately share information about similarities between data points and computing partial sums over large datasets. It uses a clever technique called input perturbations that adds noise to the data and then adjusts it back to make sure it’s still useful. This helps protect people’s privacy while still allowing us to learn important insights from the data. |
