BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1725
DTSTAMP:20260531T182256Z
SUMMARY:Synthesizing POMDP Policies: Sampling Meets Model-checking via Lear
 ning
DESCRIPTION:Speaker: Anirban Majumdar (TIFR)\n\nAbstract: \nPartially Obser
 vable Markov Decision Processes (POMDPs) are the standard framework for de
 cision-making under uncertainty. While sampling-based methods scale well\,
  they lack formal correctness guarantees\, making them unsuitable for safe
 ty-critical applications. Conversely\, formal synthesis techniques provide
  correctness-by-construction but often struggle with scalability\, as gene
 ral POMDP synthesis is undecidable. To bridge this gap\, we propose a synt
 hesis framework that integrates sampling\, automata learning\, and model-c
 hecking. Inspired by Angluin's $L^*$ algorithm\, our approach utilizes sam
 pling as a membership oracle and model-checking as an equivalence oracle. 
 This enables the synthesis of finite-state controllers with formal guarant
 ees\, provided the sampling-induced policy is regular. We establish a rela
 tive completeness result for this framework. Our experimental results demo
 nstrate that this method successfully solves threshold-safety problems tha
 t remain challenging for existing formal synthesis tools.\n \nThis is a j
 oint work with Debraj Chakraborty\, Sayan Mukherjee\, Prince Mathew and Je
 an-François Raskin.\n
URL:https://www.tcs.tifr.res.in/web/events/1725
DTSTART;TZID=Asia/Kolkata:20260602T160000
DTEND;TZID=Asia/Kolkata:20260602T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
END:VCALENDAR