Feb. 6, 2024, 5:11 a.m. | Mark Bun Gautam Kamath Argyris Mouzakis Vikrant Singhal

cs.CR updates on arXiv.org arxiv.org

We give an example of a class of distributions that is learnable in total variation distance with a finite number of samples, but not learnable under $(\varepsilon, \delta)$-differential privacy. This refutes a conjecture of Ashtiani.

class cs.cr cs.ds delta differential privacy distribution distributions privacy privately stat.ml under

SOC Manager

@ Medibank | DOCKLANDS, VIC, AU, 3008

Mobile Developer (Cortex XDR)

@ Palo Alto Networks | Tel Aviv-Yafo, Israel

IT Security Manager

@ Chubb | Hong Kong

Senior Consultant (Multiple Positions Available)

@ Atos | Plano, TX, US, 75093

Consultant (Multiple Positions Available)

@ Atos | Plano, TX, US, 75093

Information Technology Senior Consultant

@ Dezign Concepts LLC | Fort Meade, MD