Chandrajit Bajaj, docente da Universidade do Texas, apresenta a palestra “Beyond Digital Twins: Learned Dynamic Forms that Auto-Reconfigure to Function Better”.
A comunicação decorre na Sala de Reuniões do Departamento de Informática, às 14h30 no dia 15 de setembro (segunda-feira).
Beyond Digital Twins: Learned Dynamic Forms that Auto-Reconfigure to Function Better
Chandrajit Bajaj
Department of Computer Science, Mathematics and Oden Institute
Center for Computational Visualization
University of Texas at Austin
Improvement in many kinds of domains and industry are currently heavily dependent on continual discovery of newer forms with improved function aka with enhanced dynamic properties. For example, semiconductor and battery technology used in everyday electronics (e.g. mobile phones, laptops, autonomous vehicles) heavily depends on the progressive development of improved semiconductors, and electro-photonic materials. The same is true for newer drugs for improved therapeutics and better patient healthcare with progressively improved interventions. The core of our methodology is a statistical physics approach using reinforcement learned stochastic Hamiltonians with optimal control, that continually discovers new dynamic digital forms of enhanced function. The progressive stochastic Hamiltonians model dynamic geometric and topological changes along task optimized pathways in the environment , along which they perform their dynamical function. In this talk I shall provide details of this novel differential reinforcement learning framework, with continuous state and action spaces that simultaneously explores and exploits forward and inverse problems, using symplectic and dissipative stochastic processes.