BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
BEGIN:VEVENT
UID:672@lincs.fr
DTSTART;TZID=Europe/Paris:20211109T100000
DTEND;TZID=Europe/Paris:20211109T130000
DTSTAMP:20211104T144509Z
URL:https://www.lincs.fr/events/phd-thesis-defense-stochastic-matching-mod
 els-and-their-applications-to-demand-supply-balancing/
SUMMARY:PhD thesis defense "Stochastic matching models and their
 applications to demand-supply balancing"
DESCRIPTION:The recent growth of the collaborative economy with
 peer-to-peer networks created the need for interfaces that put in relation
 different types of populations. Applications such as ride sharing\, house
 renting\, job search\, dating websites and so on\, link people together
 based on various preferences or compatibilities. This motivates the
 development of new algorithms that optimize those associations\, also
 called matchings. Matching models have been studied at great length in the
 literature when parts of the population is static. However\, when customers
 arrive randomly over time\, the problem is much more challenging and many
 questions remain open. This thesis tackles the matter through the lens of
 control in stochastic matching models.\nThese models are formalized as
 follows. Items of different classes arrive to the system according to a
 given probability distribution. Upon arrival\, each item is matched with
 another compatible item according to a matching policy and both items leave
 the system immediately. If there are no compatible items\, the new arrivals
 join the queues of unmatched items of the same class. Compatibilities
 between item classes are defined by a connected graph\, where nodes
 represent the classes of items and the edges the compatibilities between
 item classes. Unmatched items induce a holding cost at each time step and
 we model this problem as a Markov Decision Process.\nIn particular\, we
 study the Bipartite Matching Model which considers bipartite compatibility
 graphs and arrivals by pair. We are interested in finding the optimal
 control for the discounted and average cost problems. For the N-shaped
 compatibility graph\, we fully characterize the optimal matching policy. We
 also identify optimal matchings for more general bipartite graphs under
 additional assumptions on the costs.\nWe also analyze the performance of
 the General Matching Model\, where new items arrive one by one with a
 non-bipartite compatibility graph\, under the First Come First Matched
 policy. We show that such a model may exhibit a non intuitive behavior:
 increasing the matching flexibility by adding new edges in the
 compatibility graph may lead\, when the system is close to the stability
 condition\, to a larger average population at the steady state.\nFinally\,
 we use Reinforcement Learning to find the optimal threshold-type policy for
 Bipartite Matching models. We develop an Actor-Critic algorithm composed of
 a Temporal Difference critic and a Policy Gradient actor that includes this
 threshold type structure and learns the optimal threshold within this set
 of policies.\n&nbsp\;\nKeywords: Stochastic matching models\, Markov
 decision processes\, Optimal control\, Braess paradox\, Reinforcement
 learning\n&nbsp\;\nHere the link to
 attend:&nbsp\;https://global.gotomeeting.com/join/212030933
CATEGORIES:PhD Defense
LOCATION:LINCS Seminars room\, 23\, avenue d'Italie\, Paris\, 75013\,
 France
GEO:48.828400;2.356897
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=23\, avenue d'Italie\,
 Paris\, 75013\, France;X-APPLE-RADIUS=100;X-TITLE=LINCS Seminars
 room:geo:48.828400,2.356897
END:VEVENT
BEGIN:VTIMEZONE
TZID:Europe/Paris
X-LIC-LOCATION:Europe/Paris
BEGIN:STANDARD
DTSTART:20211031T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
END:VTIMEZONE
END:VCALENDAR