February 2nd, 2012 @ 11:30am Chaitu Ekanadham (NYU)

Continuous basis pursuit and application to neural spike identification

Several methods decompose signals into a linear combination of a few atomic features chosen from a large dictionary, which is either constructed by hand or learned from data. However, many real signals are generated by transformation-invariant processes. For instance, an edge can occur in an image at various spatial locations, scales and orientations. An acoustic feature can occur at various times and frequencies. In this talk I will introduce Continuous Basis Pursuit, a method for inferring sparse decompositions of signals that account for continuous transformations. I will demonstrate its advantage over standard sparse decomposition methods for 1-D translation-invariant signals, and also describe its application to the “spike sorting” problem in neuroscience.

Relevant papers

C Ekanadham, D Tranchina and E P Simoncelli. A blind sparse deconvolution method for neural spike identification. Adv. Neural Information Processing Systems (NIPS*11), vol.24 2011.

C Ekanadham, D Tranchina and E P Simoncelli. Recovery of sparse translation-invariant signals with continuous basis pursuit.. IEEE Trans Signal Processing, vol.59(10), pp. 4735–4744, Oct 2011.

seminars/seminaritems/2012-02-02.txt · Last modified: 2012/01/24 23:49 by silberman