Meeting Banner
Abstract #3259

Multitask Machine Learning for Brain-State Classification

Yash S. Shah1, Ashish Farmer1, Luis Hernandez-Garcia1, Douglas C. Noll1, Mark Greenwald2, Jon-Kar Zubieta1, Scott J. Peltier1

1University of Michigan, Ann Arbor, MI, United States; 2Wayne State University, Detroit, MI, United States

Multitask learning formulation presents a novel way of accommodating information from other subjects' data and building a generalized classifier. In our study, we use multitask learning to classify the temporal crave-state of a nicotine dependent subject and compare results to standard single subject SVM. We demonstrate that multitask learning is a promising novel analysis technique for fMRI data analysis.

Keywords

abstinence accordance accurate acquisition activation allow among applications axial better block blocks bold brain build building cast choice class classes classification classifier classifiers classifying closely common comp considered coupled craving daily decision define defined denote dependence dependent depicted described discriminate discriminatory emotional employing ensure entire equals especially ever every expired extent farmer feature features finding fixation formulation function functional generalized gilbert good helps hence highest implies included increasingly independently individual induce influences intensities iterations kernel king lambda learn learning linear machine majority makes making mark matrix measuring model models newer nicotine noteworthy optimal otherwise overnight pair paradigm patterns performance picked picking predict previously problem randomly ranked real recent reduced reduction related relations response samples scanned scenes score scores screening selected separate series several shah similarity simultaneously slices smoked smoking studies subject subjects subset table take taking task tasks together trained training univariate unseen users validation variance varying verified volume volumes whereas