The series hosts a seminar every other week on current research topics. The seminar often features an invited guest speaker and occasionally local faculty members, students or others affiliated with the department. The usual time of the seminar is 3:30-4:30 pm on Fridays. Professors  Tatiyana V Apanasovich (, Qing Pan ( and Emre Barut ( ) are the Seminar Series Coordinators.

Department Seminars in Fall 2014

Upcoming Seminar

Date: Friday, Sep 12, 3:30-4:30pm

Location: Duques Hall, Room 251

Title: Network Inference from Grouped Data

Speaker: Dr. Yunpeng Zhao, George Mason University

Abstract: Within the field of network analysis, there is often an important distinction made between physical networks (i.e. highways, router systems, and electrical grids) and social networks (i.e. friendships, movie actors, and citations). In physical networks, the network topology is observable and analysis of the network properties directly informs the means by which the network functions. However, in social networks, the network topology is not explicit and must be inferred from the observed behavior. This effort is often complicated by the use of heuristic techniques for network inference which are not capable of reproducing the original behavior. In this presentation, the authors define a network based model to describe the social grouping behavior and present a maximum likelihood technique for inferring the network most likely to have produced the observed behavior.

Next Seminar

Date: Friday, Sep 26, 3:30-4:30pm

Location: Duques Hall, Room 251

Title: Optimal sample size for ethical multi-stage clinical trials

Speaker: Dr. Philippe Rigollet, Princeton University

Abstract: Ethical clinical trials, where the goals is to administer the best of two treatments to the most patients, was the original motivation behind multi-armed bandits when Thompson introduced this problem 1933. Since then, many variants of this problem have been investigated, especially under the impetus of online advertising.While in the traditional bandit setup, one is allowed to reevaluate the allocation strategy after each patient, we study a framework where the strategy must function is a small number of stages, typically 2 or 3. Such a restriction is particularly resounding when the treatment effect can only be measured days or weeks after administration. Our minimax analysis provides guidelines for the size of the stages as well as the allocation policy within each stage. Moreover, we show that a very small number of stages (at most 5) is already enough to recover the optimal bounds from the unrestricted setup.[Joint work with Sylvain Chassang (Princeton), Vianney Perchet (Paris 7) and Erik Snowberg (Caltech