Seminar by Dr. Abbas Mehrabian (McGill University)

Speaker: Dr. Abbas Mehrabian (McGill University)                                                                                                                  

Title:  New techniques for distribution learning

Date: Wednesday January 31, 2018

Time: 10.00am-11:30am

Room: EV 1.162

ABSTRACT

The general goal of unsupervised learning is to find structure in the data. Distribution learning (also known as density estimation) is the task of explicitly estimating the distribution underlying the data, which can then be explored to find structure in the data, or to generate new data. Suppose we are given a sample generated from an unknown target distribution, and want to output a distribution that is close to the target. What is the smallest needed sample size to guarantee successful learning? We introduce new techniques for mathematically bounding the sample complexity for learning continuous distributions, focusing on the class of mixtures of high-dimensional Gaussians. Joint work with Hassan Ashtiani and Shai Ben-David.

BIO

 After graduating from Waterloo in 2015, Abbas visited the Simon Fraser University, the University of British Columbia, and the University of California at Berkeley. He is now a CRM-ISM postdoctoral fellow at McGill University. He works on theoretical machine learning, randomized algorithms, and mathematical foundations of network science.