Linear-Quadratic regulators are canonical models in sequential decision-making. In adaptive control, both the stochastic linear dynamics of the state vectors as well as the effect of control inputs are unknown. The goal is to effectively learn from state observations to efficiently minimize a quadratic cost function of state and input. For control under uncertainty, decision-making algorithms are desired to stabilize the system and estimate optimal adaptive policies as quickly as possible. For this purpose, the first challenge is design and analysis under system instability. Further challenges consist of balancing the trade-off between control and identification. We present easily implementable algorithms for accurate estimation and optimal actuation at the same time and provide novel analyses for establishing performance guarantees.
Bio: Mohamad Kazem Shirani Faradonbeh is an assistant professor of Data Science in the Department of Statistics at the University of Georgia. During Fall 2020, he was a fellow of Theory of Reinforcement Learning program in Simons Institute for the Theory of Computing at the University of California - Berkeley. From 2017 to 2020 he was a postdoctoral research associate with the Informatics Institute and with the Department of Statistics at the University of Florida. He received his PhD degree in statistics from the University of Michigan, Ann Arbor in 2017, and his BSc degree in electrical engineering from Sharif University of Technology, Tehran, Iran, in 2012.