Paper privately selects clipping bounds for heavy-tailed data
Original title: Privately Clipping Heavy-Tailed Data
The method picks a clipping bound by balancing residual coherence against the extra noise variance.
A differential-privacy paper addresses privately choosing a clipping bound before summing vectors and adding noise. The method optimizes residual coherence, the directional alignment of vectors above the bound, against the variance cost of additional noise. The authors prove a close link between residual coherence and bias for a general class of heavy-tailed regimes and report that the method outperforms existing baselines across several datasets.
Unconfirmed: Dataset names and the size of the reported gains are not stated in the abstract.
- Standard private sums clip vectors, sum them, and add noise scaled to the bound.
- The paper says few tools privately choose that bound from the data.
- Empirical results claim better performance than existing baselines on several datasets.
From the original article
1 min read at arxiv.orgSumming vectors is a basic task in differentially private data analysis. Standard private sum algorithms clip vectors to some bound, sum them, and add noise scaled to the bound. The differential privacy literature has developed a deep library of noise addition mechanisms, but it offers few tools for privately choosing a clipping bound based on the data.
By Matthew Joseph · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.