Centers & Programs

Seminars

Home Centers & Programs AI and Natural Sciences Seminars

FIELD
AI and Natural Sciences
DATE
Sep 30 (Mon), 2024
TIME
13:30 ~ 15:30
PLACE
7323
SPEAKER
노승문
HOST
Choi, Jaewoong
INSTITUTE
연세대학교
TITLE
Rate-Distortion Theoretic Approach in Privacy-Constrained Mean Estimation
ABSTRACT
In this talk, I will introduce rate-distortion theoretic approaches to differentially private mean estimation. Regarding the rate- privacy-utility tradeoff in private mean estimation, we advance the characterization of an exactly optimal approach under shared randomness—a random variable shared between server and user—and identify several conditions for exact optimality, including the use of a rotationally symmetric shared random codebook. We also propose a novel randomization mechanism employing a randomly rotated simplex as the codebook, which meets the criteria for an exact-optimal codebook.
FILE