Robustness vs Efficiency in High-Dimensional Unsupervised Learning

Speaker: Ilias Diakonikolas , University of Southern California

Date: Tuesday, May 16, 2017

Time: 4:00 PM to 5:00 PM Note: all times are in the Eastern Time Zone

Refreshments: 3:45 PM

Public: Yes

Location: G449 Patil/Kiva

Event Type:

Room Description:

Host: Aleksander Madry, Ankur Moitra, Vinod Vaikuntanathan, Virginia Vassilevska Williams, MIT

Contact: Patrice Macaluso, 617-253-3037,

Relevant URL:

Speaker URL: None

Speaker Photo:

Reminders to:,,

Reminder Subject: TALK: Robustness vs Efficiency in High-Dimensional Unsupervised Learning

Consider the following basic problem: Given corrupted samples from a high-dimensional Gaussian, can we efficiently learn its parameters?
This is the prototypical problem in robust statistics, a field that took shape in the 1960's with the pioneering works of Tukey and Huber. Unfortunately, all known robust estimators are hard to compute in high dimensions. This prompts
the following question: Can we reconcile robustness and computational efficiency in high-dimensional unsupervised learning?

I will start by reviewing recent algorithmic work, giving the first efficient robust estimators in high dimensions that are able to tolerate a constant fraction of corruptions. In the main part of the talk, I will describe a range of fundamental high-dimensional estimation tasks where robustness creates computational
and/or statistical barriers.

(The main part of the talk will be based on joint work with Daniel Kane (UCSD) and Alistair Stewart (USC).)

Short bio: Ilias Diakonikolas is an Assistant Professor and Andrew and Erna Viterbi Early Career Chair in the Department of Computer Science at the University of Southern California. He is also a member of the CS Theory group and the Machine Learning center. Prior to joining USC's faculty, Ilias spent two years at UC Berkeley as the Simons Postdoctoral Fellow in Theoretical Computer Science, and then was a faculty member at the University of Edinburgh. Ilias obtained his Ph.D. in Computer Science at Columbia University, advised by Mihalis Yannakakis. Before that, he did his undergraduate studies in Greece, at the National Technical University of Athens

Research Areas:

Impact Areas:

This event is not part of a series.

Created by Patrice Macaluso Email at Monday, May 15, 2017 at 3:52 PM.