Learning Solution Map of Controlled Dynamical Systems
Abstract: In this talk, the problem of continuous-time learning of the solution map of a nonlinear dynamical system with inputs will be considered. In particular, the focus will be on the systems that exhibit non-fading memory behavior, i.e., input has a long-lasting effect on the system’s output. It will be shown that the mechanism of feedback is at the core of learning such a system. Furthermore, only the discrete-time abstraction of the feedback is needed for continuous-time learning. The required assumptions hold for systems whose dynamics are well-posed ordinary differential equations and with practically relevant classes of input signals. This enables the use of off-the-shelf solutions for learning flow functions in continuous time from sampled trajectory data. Experimental validation is performed on the model of biological neurons which demonstrates the capabilities of the proposed architecture.
Speaker Bio: Dr. Amritam Das is an assistant professor at Eindhoven University of Technology where he is part of the Control Systems (CS) group at the Electrical Engineering (EE) department. Previously, he held post-doctoral fellowship at KTH Royal Institute of Technology, Sweden and the University of Cambridge, UK. He received MSc. in Systems & Control and PhD in Electrical Engineering from Eindhoven University of Technology in 2016 and 2020, respectively. His research interests are robust and nonlinear control of multi-physics systems, physics-enabled learning for control, and model reduction applied to various technology trends in high-tech systems, power generation and neuroengineering.
Speaker Bio: Dr. Amritam Das is an assistant professor at Eindhoven University of Technology where he is part of the Control Systems (CS) group at the Electrical Engineering (EE) department. Previously, he held post-doctoral fellowship at KTH Royal Institute of Technology, Sweden and the University of Cambridge, UK. He received MSc. in Systems & Control and PhD in Electrical Engineering from Eindhoven University of Technology in 2016 and 2020, respectively. His research interests are robust and nonlinear control of multi-physics systems, physics-enabled learning for control, and model reduction applied to various technology trends in high-tech systems, power generation and neuroengineering.
