Or Sheffet: "Utilitarian Models of Privacy-Loss and Social Choice"

Date: 

Monday, September 29, 2014, 11:30am to 1:00pm

Location: 

Maxwell Dworkin 119

CRCS Lunch Seminar

Date: Monday, September 29, 2014
Time: 11:30am – 1:00pm
Place: 33 Oxford St., Maxwell Dworkin 119

Speaker: Or Sheffet, Postdoctoral Fellow, CRCS, SEAS, Harvard University.

Title: Utilitarian Models of Privacy-Loss and Social Choice


Abstract:
This talk surveys two independent works. In its first part we discuss the problem of analyzing the effect of privacy concerns on the behavior of selfish and utility-maximizing agents. Previous works [GR11, Xiao13, NOS12, CCKMV13] avoid the need to provide an explicit formalization of privacy concerns by designing mechanisms that adhere to the worst-case notion of differential privacy. Our work takes the complimentary approach and is aimed at a better understanding of the behavior of agents when the privacy concerns are explicitly formalized. Specifically we characterize the behavior of selfish utility-maximizing agents in a toy-setting where agent A's incentive to discover agent B's secret type is the result of some payments between B and A.

In the second part of the talk we discuss the problem of Social Choice, where n individuals are picking together one alternative out of m possible alternatives using a social choice function -- a function that takes as input the n individuals' preferences among the alternatives and outputs a single chosen alternative, called the winner. Inspired but newly formulated ideas as to the role of clustering [BBG09], we view social choice as a proxy for maximizing social welfare.  Our premise is that agents have (possibly implicit or latent) utility functions, and the goal of a social choice function is to maximize the social welfare — i.e., (possibly weighted) sum of agent utilities — of the selected alternative. We will also discuss current, open ended, work as to maximizing utility of a matching / stable matching.

Based Joint work with Yiling Chen and Salil Vadhan (WINE'14) and Craig Boutillier, Ioannis Caragiannis, Simi Haber, Tyler Lu and Ariel Procaccia (EC' 12).

Biography:
I am a fellow of Harvard's CRCS and a member of the Privacy-Tools program. Prior to my post-doc at Harvard I was a research fellow in the Simon's Institute for the Theory of Computer Science as a member of the "Theoretical Foundations of Big-Data" program. I have a B.SC in math and computer science from the Hebrew University in Jerusalem, Israel, and a M.Sc in computer science and applied math from the Weizmann Institute of Science. I have a PhD in computer science from Carnegie Mellon University, where I had the honor of being advised by prof. Avrim Blum. My research interests lie in differential privacy, algorithmic game theory, machine learning in general and clustering in particular.