Technion, IEM faculty - Operations Research & Statistics seminar
 
Speaker: Dmitri "Mitya" Chklovskii, Janelia Farm, HHMI
 
Title: Can neurons be modeled by online learning algorithms?
 
Date: 24/12/2012
 
Time: 12:30
 
Place: Bloomfield-527
 
Abstract:  <http://ie.technion.ac.il/seminar_files/1355819038_20121224.pdf>
 
Or read it here:
 
Abstract:
The astounding complexity of human thoughts, feelings, and behavior is
generated by a self-organizing network of 10^11 neurons. What basic com-
putation is performed by each neuron in such a network? We propose to
model neuronal computation by online learning algorithms for estimation
and prediction of time varying signals. First, by using projection pursuit,
synaptic learning rules could discover a linear combination of inputs with
non-Gaussian, e.g. Laplace distribution. Second, Laplacian signal contami-
nated by Gaussian noise can be robustly estimated by an integrator followed
by a threshold linear response function common firing rate model of a neu-
ron. If the estimation signal must be encoded by a binary sequence, the
estimator reduces to an integrate-and-fire neuron
 
---------------------------------------------------------
Technion Math. Net (TECHMATH)
Editor: Michael Cwikel   <techm@math.technion.ac.il> 
Announcement from:  <tomerk@.technion.ac.il>