BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:EESS talk on "Bayesian Parameter Inference of Nonlinear Stochastic
  Differential Equation Models"
DTSTART:20160301T121500
DTEND:20160301T131500
DTSTAMP:20260510T053501Z
UID:201308f8f8a2bb6d8bc74a081ffb5da7bbad408311056fc9a37f9d1f
CATEGORIES:Conferences - Seminars
DESCRIPTION:Dr Albert CARLO\, Head of Mathematical Methods in Env. Researc
 h\, Dept of Systems Analysis\, Integrated Assessment and Modelling\, EAWAG
 \, CH\nAbstract:\nParameter inference is a fundamental problem in data-dri
 ven modeling. Given observed data\, which is believed to be a realization 
 of some parameterized model the aim is to find parameter values that are a
 ble to explain the observed data. If the model is to make reliable predict
 ions it must contain the dominant sources of uncertainty\, which naturally
  leads us to stochastic models. Stochastic models render parameter inferen
 ce much harder\, as its aim is to find the whole distribution of likely pa
 rameter values. In Bayesian statistics\, which is a consistent framework f
 or data-driven learning\, this so-called posterior distribution expresses 
 the probability of a given parameter to be the "true" one\, and can be use
 d to make probabilistic predictions.\nFor truly stochastic models this pos
 terior distribution is typically extremely expensive to evaluate. There is
  a number of sampling techniques\, for Bayesian parameter inference\, that
  avoid repeated evaluation of the posterior. Sequential filtering methods 
 (Kalman filters\, particle filters) form a class of powerful methods. Appr
 oximate Bayesian computation (ABC) methods are of particular importance if
  the data is rare or expensive.\nIn my talk\, I propose a novel\, exact an
 d very efficient\, approach for generating posterior parameter distributio
 ns\, for stochastic differential equation models calibrated to measured ti
 me-series. The algorithm is inspired by re-interpreting the posterior dist
 ribution as a statistical mechanics partition function of an object akin t
 o a polymer\, whose dynamics is confined by both the model and the measure
 ments. To arrive at distribution samples\, we employ a Hamiltonian Monte C
 arlo approach combined with a multiple time-scale integration. A separatio
 n of time scales naturally arises if either the number of measurement poin
 ts or the number of simulation points becomes large. Furthermore\, at leas
 t for 1D problems\, we can decouple the harmonic modes between measurement
  points and solve the fastest part of their dynamics analytically.  Our a
 pproach is applicable to a wide range of inference problems and is highly 
 parallelizable.Short biography:\nCarlo Albert was born in Valais\, Switzer
 land. He studied mathematics at ETHZ and did his PhD in Theoretical Physic
 s with Prof. Jürg Fröhlich. After a short Post Doc at the mathematics de
 partment of the University of Geneva he moved to Eawag\, the aquatic resea
 rch institute of the ETH domain. Since autumn 2012 he holds a tenure track
  position and builds up a group for mathematical methods in the environmen
 tal sciences.
LOCATION:GR C0 01 http://plan.epfl.ch/?room=GR%20C0%2001
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR