Discovering a Semantic Basis of Neural Activity Using Simultaneous Sparse Approximation - Robotics Institute Carnegie Mellon University

Discovering a Semantic Basis of Neural Activity Using Simultaneous Sparse Approximation

Workshop Paper, ICML '08 Sparse Optimization and Variable Selection Workshop, July, 2008

Abstract

We consider the problem of predicting brain activation in response to arbitrary words in English. Whereas previous computational models have encoded words using predefined sets of features, we formulate a model that can automatically learn features directly from data. We show that our model reduces to a simultaneous sparse approximation problem and show two examples where learned features give insight about how the brain represents meanings of words.

BibTeX

@workshop{Palatucci-2008-10050,
author = {Mark Palatucci and Tom Mitchell and Han Liu},
title = {Discovering a Semantic Basis of Neural Activity Using Simultaneous Sparse Approximation},
booktitle = {Proceedings of ICML '08 Sparse Optimization and Variable Selection Workshop},
year = {2008},
month = {July},
keywords = {sparse approximation, fMRI},
}