Algorithms and Complexity Seminar: 2013-2014: Robust Bayesian Inference
, UC Berkeley
Date: Wednesday, March 19, 2014
Time: 4:00 PM to 5:00 PM Note: all times are in the Eastern Time Zone
Host: Ilya Razenshteyn and Henry Yuen
Contact: Patrice Macaluso, email@example.com
Relevant URL: http://www.ilyaraz.org/acseminar/
Speaker URL: None
TALK: Algorithms and Complexity Seminar: 2013-2014: Robust Bayesian Inference
We show a near optimal policy for Bayesian inference when an adversary can modify inputs using one of a known set of arbitrary modification rules. Given a black-box access to a Bayesian inference in the classic (adversary-free) setting, our near optimal policy runs in polynomial time in the number of inputs and the number of modification rules. This allows to handle the interesting case where the adversary controls a constant number of inputs. For the special case of aggregating noisy signals from independent experts, We address the case where the expert's accuracy is unknown, and show that our policy applies to this case as well.
Joint work with Yishay Mansour and Moshe Tennenholtz.
Created by Patrice Macaluso at Tuesday, March 18, 2014 at 4:25 PM.