Graphical models for sparse estimation and neuroscience

Start: 01/26/2012 - 4:15pm
End  : 01/26/2012 - 5:15pm

Applied Math Seminar

Alyson Fletcher (University of California, Berkeley)


Many signals are sparse or have far fewer degrees of freedom than their high-dimensional ambient space.  The prominence of sparsity in the natural world, coupled with the prospect of increasingly large modern data sets, drives the question of how low-dimensional structures can be efficiently inferred from data.  Such algorithms are typically nonlinear and difficult to analyze.  In this talk, I will present a new systematic low-complexity framework for sparse estimation based on approximate loopy belief propagation in graphical models.  I will then address estimation problems in neuroscience: neural connectivity mapping and visual receptive field estimation.  Promising future work includes neurological modeling of epilepsy seizures and image recovery in multi-coil MRI.


Millikan 134, Pomona College

Misc. Information

Refreshments at 3:45 in CGU Math South.
Wine and cheese after the talk, same place.