Probabilistic methods in dynamics

Machine Learning of Chaotic Dynamical Systems

speaker: Hongkun Zhang (University of Massachusetts Amherst)

abstract: In this talk, I present a novel approach to predict the behavior of chaotic dynamical systems using a newly designed neural network Discrete-Temporal Sobolev Networks (DTSN). Chaotic systems are notoriously difficult to model due to their inherent sensitivity to initial conditions and unpredictable long-term behavior. Our recent work focuses on two well-known chaotic systems: the Chirikov Standard Map and the Lorentz System. By harnessing the power of DTSN, we demonstrate the potential for machine learning to effectively learn and predict the complex dynamics of these systems. We evaluate the performance of our models using various metrics and compare them to traditional numerical integration methods. Our results show that DTSN networks can effectively learn the underlying dynamical structure of these chaotic systems. We also discuss potential extensions of our work, such as incorporating the Transformer Neural Networkf or addressing different types of chaotic systems.

timetable:
Wed 31 May, 11:00 - 11:50, Aula Dini
<< Go back