Propose-filter-release turns friendly privacy into standard differential privacy
Original title: Propose-Filter-Release: Scaling Noise to Typical Sensitivity via Private Filtering
PFR privately filters outliers so noise can scale to typical sensitivity, covering statistics, subspaces, graphs and treatment rules.
The paper introduces propose-filter-release (PFR), which privately filters records so a mechanism that is private for typical additions or removals becomes differentially private on every input. It needs a concave friendliness score and a friendly-DP release mechanism, with privacy guaranteed under public conditions on the number of records. The authors present it as a general primitive comparable to propose-test-release, with FriendlyCore as one implementation, and instantiate it for scalar statistics, principal subspaces, histogram KL divergence, graph effective resistances and treatment-rule selection.
- Noise is scaled to typical rather than worst-case single-record influence.
- The filtering step itself must be private.
- FriendlyCore is described as one implementation of PFR.
- Instantiations include scalar statistics and graph effective resistances.
From the original article
1 min read at arxiv.orgAn estimator can be accurate and stable on typical data, yet sensitive to a few possible records. Differentially private mechanisms commonly scale their noise to the worst-case influence of a single record, which lets these records dictate the noise even when the remaining data would allow far less. Removing them first would restore the smaller scale, but the removal must itself be private.
By Moshe Shenfeld · 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.