Section: New Results

Semi-Supervised Factored Logistic Regression for High-Dimensional Neuroimaging Data

Imaging neuroscience links human behavior to aspects of brain biology in ever-increasing datasets. Existing neuroimaging methods typically perform either discovery of unknown neural structure or testing of neural structure associated with mental tasks. However, testing hypotheses on the neural correlates underlying larger sets of mental tasks necessitates adequate representations for the observations. We therefore propose to blend representation modelling and task classification into a unified statistical learning problem. A multinomial logistic regression is introduced that is constrained by factored coefficients and coupled with an au-toencoder. We show that this approach yields more accurate and interpretable neural models of psychological tasks in a reference dataset, as well as better generalization to other datasets.

Figure 3. Classification weight maps. The voxel predictors corresponding to 2 exemplary (of 18 total) psychological tasks (rows) from the Human Connectome Project dataset. Left column: multinomial logistic re- gression (same implementation but without bottleneck or autoencoder), middle column: Semi-Supervised Factored Logistic Regression (SSFLogReg), right column: voxel-wise average across all samples of whole-brain activity maps from each task. SSFLogReg puts higher absolute weights on relevant structure, lowers ones on irrelevant structure, and yields BOLD-typical local con- tiguity (without enforcing an explicit spatial prior). More information can be found in [50] .

More information can be found in [50] .