BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Computational Neuroscience Seminars: Spike-Based Probabilistic Inf
 erence in Analog Graphical Models Using Interspike-Interval Coding
DTSTART:20130905T133000
DTSTAMP:20260407T025855Z
UID:cf8964b1b9f0e745246758e9533ec49b687c4f7c80ff3aae31bd7528
CATEGORIES:Conferences - Seminars
DESCRIPTION:Andreas STEIMER\, Institute of Neuroinformatics\, Uni Zurich\n
 Temporal spike codes play a crucial role in neural information processing.
 \nIn particular\, there is strong experimental evidence that interspike in
 tervals (ISIs) are used for stimulus representation in neural systems.\nHo
 wever\, very few algorithmic principles exploit the benefits of such tempo
 ral codes for probabilistic inference of stimuli or decisions. In this tal
 k I will describe the functional properties of a spike-based processor tha
 t uses ISI distributions to perform probabilistic inference.\nThe abstract
  processor architecture serves as a building block for more concrete\, neu
 ral implementations of the belief-propagation (BP) algorithm in arbitrary 
 graphical models (e.g.\, Bayesian networks and factor graphs). The distrib
 uted nature of graphical models matches well with the architectural and fu
 nctional constraints imposed by biology. In our model\, ISI distributions 
 represent the BP messages exchanged between factor nodes\, leading to the 
 interpretation of a single spike as a random sample that follows such a di
 stribution. I will show simulation results that verify the functionality o
 f the abstract processor model in full graphs\, and demonstrate that it ca
 n be applied even in the presence of analog variables. As a particular exa
 mple\, I will also show results of a concrete\, neural implementation of t
 he processor\, although in principle our approach is more flexible and all
 ows different neurobiological interpretations. Furthermore\, electrophysio
 logical data from area LIP during behavioral experiments are assessed in l
 ight of ISI coding\, leading to concrete testable\, quantitative predictio
 ns and a more accurate description of these data compared to hitherto exis
 ting models.
LOCATION:BC 02 https://plan.epfl.ch/?room==BC%2002
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR