BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Dr. Qi Tang
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260513T222325Z
UID:tag:localist.com\,2008:EventInstance_47913118706252
DTSTART:20241107T161500Z
DTEND:20241107T171500Z
DESCRIPTION:Title: Leveraging low-dimensional structures in structure-prese
 rving machine learning for dynamical systems\n\nAffilication: Georgia Inst
 itute of Technology\n\nAbstract: \n\nIn this talk I will discuss our recen
 t effort to develop structure-preserving machine learning (ML) for time se
 ries data\, focusing on both dissipative PDEs and singularly perturbed ODE
 s.\n\n \n\nThe first part presents a data-driven modeling method that accu
 rately captures shocks and chaotic dynamics through a stabilized neural OD
 E framework. We learn the right-hand-side of an ODE by adding the outputs 
 of two networks together\, one learning a linear term and the other a nonl
 inear term. The architecture is inspired by the inertial manifold theorem.
  We apply this method to chaotic trajectories of the Kuramoto-Sivashinsky 
 equation\, where our model keeps long-term trajectories on the attractor a
 nd remains robust to noisy initial conditions.\n\n \n\nThe second part exp
 lores structure-preserving ML for singularly perturbed dynamical systems. 
 A powerful tool to address these systems is the Fenichel normal form\, whi
 ch significantly simplifies fast dynamics near slow manifolds. I will disc
 uss a novel realization of this concept using ML. Specifically\, a fast-sl
 ow neural network (FSNN) is proposed\, enforcing the existence of a traina
 ble\, attractive invariant slow manifold as a hard constraint. To illustra
 te the power of FSNN\, I will show a fusion-motivated example where tradit
 ional numerical integrators all fail.
GEO:39.681267;-75.754997
LOCATION:Ewing Hall\, 336
SUMMARY:Dr. Qi Tang
URL;VALUE=URI:https://events.udel.edu/event/dr-qi-tang
CATEGORIES:Academics
CATEGORIES:College of Arts and Sciences
END:VEVENT
END:VCALENDAR