Dr. Jalal Arabneydi

In recent years, deep learning and deep reinforcement learning have been astonishingly successful  in outperforming classical results in some complex  applications such as natural language processing, image processing and Atari games. However, with all the recent progresses,  there are still some important challenges  that need to be addressed such as reliability, scalability,  explainability and robustness of the AI algorithms, to name only a few.

As an attempt to address some of the above shortcomings,  it is critical to start thinking about a new set of problems called deep planning,  where the objective is to study deep structured models parametrically (irrespective of the underlying data sets). On the other hand, deep planning is a difficult problem, in general, and that is why  it has not received much attention so far. To this end, we introduce a novel class of tractable deep planning algorithms for sequential decision making in large-scale decentralized  multi-agent systems  called deep structured team (DST). 

In DSTs, agents wish to collaborate to minimize a common cost function  while they are coupled in dynamics, cost function and information through a set of linear regressions of the states and actions of all agents. The salient feature of DSTs is  that their computational complexity is amenable to  the number of agents. In this talk, we present some theoretical results involving Markov chain models, linear quadratic regulators, Kalman filters,  reinforcement learning, model predictive control, mean-field approximation, minmax optimization, risk-sensitive cost function,  and global convergence of policy gradient methods.   In addition, we illustrate the efficacy of the theoretical results by  a few toy examples including (a)  cyber-physical attacks in swarm robotics and  (b)  integration of renewable energies in power grids, and (d) control of pandemics. If time permits, we extend DSTs  to  non-cooperative games, where  agents play a non-zero sum game.

Biography:

Dr. Jalal Arabneydi received his Ph.D. degree in Electrical and Computer Engineering from McGill University, Montreal, Canada in 2017. He was then a postdoctoral fellow at Concordia University from 2017-2021. He is currently a professional artificial intelligence scientist in JACOBB. In 2014, he was the recipient of the best student paper award at the 53rd Conference on Decision and Control (CDC). His principal research area is sequential decision making, large-scale systems,  multi-agent reinforcement learning,  and artificial intelligence.  One of his current theoretical research interests lies in the intersection of decision making theory and artificial intelligence. The ultimate goal is to define proper mathematical tools and solution concepts in order to develop large-scale decision-making algorithms that work under imperfect information and incomplete knowledge with analytical performance guarantees. To this end, he has proposed deep structured teams/games as a newly emergent class of decentrazlied large-scale planning algorithms.