ICON Seminar in Autonomy: Prof. Shreyas Sundaram (Purdue)

Time: 3:00-4:00 pm Eastern Time, February 16 (Friday), 2024

Location: MSEE 112

Zoom Link: https://purdue-edu.zoom.us/j/98798335169?pwd=ZFNMdmZXendoQ1RCRDczM2dTR1RIdz09

Free coffee and snacks will be provided.     

Resilient Distributed Coordination, Optimization, and Hypothesis Testing in Large-Scale Networks

Abstract:

A key challenge in large-scale networks is to allow each agent to learn the true state of the world, based on its own measurements and information that it obtains from others in the network.  For example, in autonomous swarms and sensor networks, what information should the agents (drones and sensors) exchange with each other in order to allow the swarm to understand the environment that it is monitoring?  Similarly, in social networks, how should individuals incorporate the opinions of their friends when updating their own beliefs about the world?  The challenge of learning in such settings is further compounded by the fact that some of the agents in the network may be malicious or adversarial, and spread misinformation in order to prevent the other agents from learning the true state.

In this talk, we will provide an overview of classes of networked dynamics (or algorithms) that have been studied over the past decade for resilient coordination and learning in networks.  We will start by describing a simple general-purpose approach to enabling resilience to adversarial nodes in several classes of distributed consensus and optimization dynamics, based on having each node remove a certain number of the most extreme values in its neighborhood at each time step.  In the process, we will describe key insights about the interplay between the network topology and the ability of the prescribed dynamics to tolerate misbehavior.  We will then describe how to build on these ideas to tackle other classes of distributed estimation problems in networks, focusing in particular on a distributed hypothesis testing setting.  We will show how such problems can be easily solved using a simple "min rule" to propagate beliefs throughout the network, and then describe how local-filtering approaches can be leveraged to provide resilience to adversarial agents in such settings.

Speaker:

Shreyas Sundaram is the Marie Gordon Professor in the Elmore Family School of Electrical and Computer Engineering at Purdue University, and a Co-Director of the Institute for Control, Optimization and Networks (ICON).  He received his Ph.D. in Electrical Engineering from the University of Illinois at Urbana-Champaign in 2009, and was a Postdoctoral Researcher at the University of Pennsylvania from 2009 to 2010.  He was an Assistant Professor at the University of Waterloo from 2010 to 2014.  His research interests include control of distributed and multi-agent systems, fault-tolerant and secure control, estimation theory, game theory, and network science. Dr. Sundaram is a recipient of the National Science Foundation CAREER award and an Air Force Research Lab Summer Faculty Fellowship.  At Purdue, he received the HKN Outstanding Professor Award, the Outstanding Mentor of Engineering Graduate Students Award, the Hesselberth Award for Teaching Excellence, and the Ruth and Joel Spira Outstanding Teacher Award.  He and his students have received several best paper honors at leading control conferences.                         

Recording:

https://www.youtube.com/watch?v=JBDkXazzRGU&ab_channel=ICONPurdue

Organizers: Ziran Wang (ziran@purdue.edu), Nak-seung Patrick Hyun (nhyun@purdue.edu), & Tho Le (thovle@purdue.edu)