BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1281
DTSTAMP:20230914T125957Z
SUMMARY:Monte Carlo tree search with advice
DESCRIPTION:Speaker: Debraj Chakraborty (Université Libre de Bruxelles)\n\
 nAbstract: \nWe show how to combine techniques from formal methods and lea
 rning for online computation of a strategy that aims at optimizing the exp
 ected long-term reward in large systems modelled as Markov decision proces
 ses. This strategy is computed with a receding horizon and using Monte Car
 lo tree search (MCTS). We augment the MCTS algorithm with the notion of ad
 vice which guides the search in the relevant part of the tree using exact 
 methods. Such advice can be symbolically written as a logical formula and 
 computed on-the-fly using model-checking tools.\n\nWe show that the classi
 cal theoretical guarantees of the Monte Carlo tree search are still mainta
 ined after this augmentation. To lower the latency of MCTS algorithms with
  advice\, we propose to replace advice coming from exact algorithms with a
 n artificial neural network. For this purpose\, we implemented an expert i
 mitation framework to train the neural network in order to replace expert 
 advice with neural advice. To demonstrate the practical interest of our te
 chniques\, we implemented the frameworks on different systems modelled as 
 MDPs.\n
URL:https://www.tcs.tifr.res.in/web/events/1281
DTSTART;TZID=Asia/Kolkata:20230321T160000
DTEND;TZID=Asia/Kolkata:20230321T170000
LOCATION:A201
END:VEVENT
END:VCALENDAR