BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning Optimal Distributionally Robust Stochastic Control in Con
 tinuous State Spaces - Jose Blanchet (Stanford University)
DTSTART:20251112T140000Z
DTEND:20251112T144000Z
UID:TALK238501@talks.cam.ac.uk
DESCRIPTION:We study data-driven learning of robust stochastic control for
  infinite-horizon systems with continuous state and action spaces. In many
  applied settings--supply chains\, finance\, manufacturing\, services\, an
 d dynamic games--the state-transition mechanism is determined by system de
 sign\, while available data capture the distributional properties of the s
 tochastic inputs from the environment. For modeling and computational trac
 tability\, a decision maker often adopts a Markov control model with i.i.d
 . environment inputs\, which can render learned policies fragile to intern
 al dependence or external perturbations. We introduce a distributionally r
 obust stochastic control paradigm that promotes policy reliability by intr
 oducing adaptive adversarial perturbations to the environment input\, whil
 e preserving the modeling\, statistical\, and computational tractability o
 f the Markovian formulation. From a modeling perspective\, we examine two 
 adversarial models--current-action-aware and current-action-unaware--leadi
 ng to distinct dynamic behaviors and robust optimal policies. From a stati
 stical learning perspective\, we characterize optimal finite-sample minima
 x rates for uniform learning of the robust value function across a continu
 um of states under ambiguity sets defined by the $f_k$-divergence and Wass
 erstein distance. To efficiently compute the optimal robust policies\, we 
 further propose algorithms inspired by deep reinforcement learning methodo
 logies. Finally\, we demonstrate the applicability of the framework to rea
 l problems.\n(Joint work with Nian Si\, Shengbo Wang\, Zhengyuan Zhou)
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR