- PRIVATE APPROXIMATIONS AND ...
- Edit Event
- Cancel Event
- Preview Reminder
- Send Reminder
- Other events happening in November 2013
PRIVATE APPROXIMATIONS AND BIG DATA
Speaker:
Kobbi Nissim
, Ben-Gurion University
Date: Friday, November 01, 2013
Time: 10:30 AM to 12:00 PM Note: all times are in the Eastern Time Zone
Public: Yes
Location: 32-G449
Event Type:
Room Description:
Host: Shafi Goldwasser
Contact: Holly A Jones, hjones01@csail.mit.edu
Relevant URL: http://toc.csail.mit.edu/node/380
Speaker URL: None
Speaker Photo:
None
Reminders to:
cis-seminars@csail.mit.edu
Reminder Subject:
TALK: PRIVATE APPROXIMATIONS AND BIG DATA
One of the main issues of big data -- scalability -- can be addressed by trading accuracy with approximation. Approximation seems hence a compelling strategy also for achieving efficiency in the context of computations performed over sensitive data, as in secure multiparty computation protocols. Care must be taken, however, so that the trading of an exact computation with an approximation would not result in leakage of more than what is intended. In 2003, Feigenbaum et al. introduced the notion of private approximation: the result of a private approximation algorithm should reveal no additional information beyond what follows from the output of the function being approximated. Subsequent work resulted in mixed results: On one hand, there exist families of problems for which private approximations enable efficient protocols. On the other hand, the approach can be shown to fail w.r.t. other problems in a very strong sense: for these problems every efficient private approximation algorithm must leak information beyond the exact computation. The latter negative results motivated looking into notions of approximation that circumvent impossibility by permitting small (quantified and well defined) leakage. In the talk, we will introduce the notion of private approximation, motivate it and examine it (and some variants) as a strategy for secure computation over large datasets. We will review the main positive and negative results, and discuss some research directions.
Research Areas:
Impact Areas:
Created by Holly A Jones at Monday, October 28, 2013 at 8:58 AM.