Latest

Every published item, in order · 7652 items
Oct 30
PapersSelected14:11
PFR Paper Introduces Private Filtering for Differential Privacy

The non-obvious point is that filtering outliers privately, instead of discarding the worst-case guarantee, can cut noise while keeping full DP — meaning better accuracy from the same privacy budget. Practitioners tuning DP deployments in statistics, graph analysis or user analytics can apply PFR to tasks where a few unusual records would otherwise dominate noise. Compared with propose-test-release, PFR needs no public stability check and covers any concave friendliness score.

arXiv cs.LG@Moshe Shenfeld et al. (3)Source
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.

Only the latest 2000 items are listed; for older ones see the daily archive