Building Nonparametric Sparse Models from Natural Images to V1 fMRI

• Activity Number: 341
• Type: Invited
• Date/Time: Tuesday, August 4, 2009 : 2:00 PM to 3:50 PM
• Sponsor: International Chinese Statistical Association
• Author(s): Bin Yu*+ and Pradeep K. Ravikumar and Vincent Vu and Thomas Naselaris and Kendrick Kay and Jack Gallant
• Companies: University of California, Berkeley and University of California, Berkeley and University of California, Berkeley and University of California, Berkeley and University of California, Berkeley and University of California, Berkeley
• Keywords: sparsity ; nonlinearity ; SpAM ; visual cortex V1 ; fMRI ; prediction

Abstract

We study the brain activity in primary cortex area V1 as measured by fMRI in collaboration with the Gallant Lab on Berkeley campus. The goal is to predict fMRI responses from the corresponding natural image stimuli, and identify potential features of the images that drive the neural activity. In this talk, I will present a nonlinear sparse model V-SpAM for each V1 voxel that is based on the SpAM model by Ravikumar et al (2007) with original and pooled image features as inputs. Then we constrain the nonlinearity to be the same for each voxel (Shared V-SpAM). Our results show that the V-SpAM and shared V-SpAM models predict fMRI responses better than the benchmark linear model. We hope that the V-SpAM framework would be particularly useful for modeling neurons or fMRI signals recorded in higher and more nonlinear stages of visual processing beyond V1.