Abstract:

Over the past twenty years, we have witnessed the rise of what has been deemed the artificial intelligence “takeover” leading to advancements from robotic vision to network multi-agent models. While significant advancements have been made, there remains an information gap to develop fully autonomous systems and at best, such frameworks require non-expert operators who possess expert knowledge to combat the risk complexities of highly uncertain environments.   During such times of duress, operators often abandon built autonomy due to a lack of trust and intractability of a well-defined (non-convex) solutions.  Whether it involves the financial crisis and the FED’s policy towards systemic banking risk, combinatoric cancer targeted drug therapy, to autonomous vehicles and robotics, there exists a need to fuse domain level experts “on-the-fly” in unison with their autonomous counterparts. Simply put, there exist no universal framework and we often utilize human operators to oversee such performance in an on-the-loop fashion to plug potential information gaps.
 
To this end, this talk will introduce an “on-the-fly” feedback control paradigm towards a broad class of optimization problems that can be found both in autonomy and network analysis.  At the mathematical core, this talk will focus the interplay of control, geometry, entropy, and optimal mass transport.  We will begin by motivating such concepts in robotic vision.  From this, we will introduce a control framework of how human interaction can provide operator input to a robotic vision-based system in a user-friendly manner.  We will then shift gears to see how geometric and control concepts can be utilized to exploit functionality of dynamic networks often found in social, biological, financial and transportation systems.  We conclude with several examples in such fields and show how interactive network control relates to notions of Maxwell’s Demon and perhaps more importantly, provides a generalized framework in which domain level experts can fuse corresponding knowledge to aid in controlling the “unknown-unknown.”

Speaker Bio:

Romeil Sandhu is currently an Assistant Professor at Stony Brook University with appointments in Computer Science, Biomedical Informatics, and Applied Mathematics & Statistics Departments and is the recipient of the 2018 AFOSR YIP Award for work on interactive feedback control for autonomous systems and 2018 NSF CAREER Award for work on geometric optimization of time-varying networks. Romeil first received his B.S. and M.S. degrees from the Georgia Institute of Technology in Electrical Engineering in 2006 and 2009, respectively. Then, under the direction of Professor Allen Tannenbaum, he completed his Ph.D. in 2011 in control-based vision and learning. Prior to his academic position at Stony Brook, he formed a startup providing government services leading to a successful exit. His academic research lies on the intersection of control, geometry, and statistics applied towards a variety of problems rooted in robotic vision, imaging, networks, systems biology, and autonomy.