BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Randomized optimization for stochastic systems: Theory and applica
 tions.
DTSTART:20091023T101500
DTSTAMP:20260406T214433Z
UID:1cb52fe3597127fa0554f03752f66dad52e7ce082dcd9dbecb8ced05
CATEGORIES:Conferences - Seminars
DESCRIPTION:Prof. J. Lygeros\, Automatic Control Laboratory (ETHZ\, Züric
 h). \nSimulated annealing\, Markov Chain Monte Carlo\, and genetic algorit
 hms are all randomized \nmethods that can be used in practice to solve (al
 beit approximately) complex optimization \nproblems. They rely on construc
 ting appropriate Markov chains\, whose stationary distribution \nconcentra
 tes on "good" parts of the parameter space (i.e. near the optimizers). Man
 y of these \nmethods come with asymptotic convergence guarantees that esta
 blish conditions under \nwhich the Markov chain converges to a globally op
 timal solution in an appropriate \nprobabilistic sense. An interesting que
 stion that is usually not covered by asymptotic \nconvergence results is t
 he rate of convergence: How long should the randomized algorithm \nbe exec
 uted to obtain a near optimal solution with high probability? Answering th
 is question \nallows one to determine a level of accuracy and confidence w
 ith which approximate \noptimality claims can be made\, as a function of t
 he amount of time available for \ncomputation. In this talk we present som
 e new results on finite sample bounds of this type\, \nprimarily in the co
 ntext of stochastic optimization with expected value criteria using Markov
  \nChain Monte Carlo methods. The discussion will be motivated by the appl
 ication of these \nmethods to collision avoidance in air traffic managemen
 t and parameter estimation for \nbiological systems.
LOCATION:MEC2405 
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR