Applying Full Correlation Matrix Analysis to Memory Data

Abstract

Full correlation matrix analysis (FCMA) is a novel technique developed by Turk-­Browne, Wang, and Singer for studying functional connectivity. It allows us to train machine learning classifiers on correlations rather than voxel intensities. In doing so, it may shed insight into communication and neural mechanisms. This thesis uses FCMA on memory data collected in the Norman lab to see if a better classifier could be trained on correlations than on intensity. The study searches through the parameter space for the optimal configuration for FCMA but fails to find a classifier with above chance accuracy using FCMA. However, the study does succeed in finding accurate classifiers using conventional intensity-­based analysis. The study goes on to speculate what this means in terms of FCMA and memory.