Sampreet Kalita, Indian Institute of Technology Guwahati, Assam, India

Library, A.2.500, Staudtstr. 2

Location details

Abstract:

Deep neural networks have been used to design and analyze several physical systems in the past decade. Whereas most works utilize the paradigms of supervised or unsupervised learning, the use of reinforcement learning (RL) to control dynamical physical systems has gathered a considerable amount of interest in recent years. Based on our recent work [arXiv:2408.12271], this talk presents a deep RL-based approach to strategically cool down coupled harmonic oscillators in an environment with high thermal noise. Following a brief introduction to the theoretical framework and the numerical implementation, we shall demonstrate the controlled cooling for multiple configurations of oscillator networks by applying RL-predicted modulated external feedback forces to specific oscillators.