BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:
X-WR-TIMEZONE:Europe/Paris
BEGIN:VTIMEZONE
TZID:Europe/Paris
X-LIC-LOCATION:Europe/Paris
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700329T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701025T030000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
SUMMARY:
Distribution-based objectives for Markov Decision Processes
STATUS:CONFIRMED
ATTENDEE;CN="Blaise Genest
":
MAILTO:no@spam.com
DESCRIPTION:
Abstract: We consider distribution-based objectives for Mark
ov Decision Processes (MDP). This class of objectives gives
rise to an interesting trade-off between full and partial in
formation. As in full observation\, the strategy in the MDP
can depend on the state of the system\, but similar to parti
al information\, the strategy needs to account for all the s
tates at the same time. In this paper\, we focus on two safe
ty problems that arise naturally in this context\, namely\,
existential and universal safety. Given an MDP A and a close
d and convex polytope H of probability distributions over th
e states of A\, the existential safety problem asks whether
there exists some distribution â in H and a strategy of A\
, such that starting from â and repeatedly applying this s
trategy keeps the distribution forever in H. The universal s
afety problem asks whether for all distributions in H\, ther
e exists such a strategy of A which keeps the distribution f
orever in H. Further\, we compare these results with existen
tial and universal safety problems for Rabinâs probabilist
ic finite-state automata (PFA)\, the subclass of Partially O
bservable MDPs which have zero observation. Compared to MDPs
\, strategies of PFAs are not state dependent. joint work wi
th S. Akshay (IIT Bombay) and Nikhil Vyas (PhD student @ MIT
).
DTSTART;TZID=Europe/Paris:20180626T110000
DURATION:PT1H
URL;VALUE=URI:http://www.lsv.ens-cachan.fr/Seminaires/?sem=201806261
100
UID:LSVsemLSV.201806261100@lsv.ens-cachan.fr
LOCATION:Salle de Conférence (Pavillon des Jardins)
END:VEVENT
END:VCALENDAR