BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning in Dynamical Systems - Elad Hazan (Princeton University)
DTSTART:20251029T130000Z
DTEND:20251029T140000Z
UID:TALK239734@talks.cam.ac.uk
CONTACT:Georg Maierhofer
DESCRIPTION:Learning in dynamical systems is a fundamental challenge under
 lying modern sequence modeling. Despite extensive study\, efficient algori
 thms with formal guarantees for general nonlinear systems have remained el
 usive. This talk presents a provably efficient framework for learning in a
 ny bounded and Lipschitz nonlinear dynamical system\, establishing the fir
 st sublinear regret guarantees in a dimension-free setting. Our approach c
 ombines Koopman lifting\, Luenberger observers\, and\, crucially\, spectra
 l filtering to show that a broad class of nonlinear dynamics are learnable
 . These insights motivate a new neural architecture\, the Spectral Transfo
 rm Unit (STU)\, which we will describe and present preliminary experiments
  on open benchmarks of language modelling and dynamical systems.
LOCATION:Centre for Mathematical Sciences\, MR14
END:VEVENT
END:VCALENDAR