Neural Information Processing Systems http://nips.cc/

To Reviewer 2/4/5/7: The neuroimaging field is still limited to *either* using unsupervised methods to find "resting-state networks" (PCA,SparsePCA,ICA,etc.) *or* using supervised methods on experimental data to predict psychological processes (logistic regression,SVMs,etc.). The unsupervised discovery of special structure and supervised prediction of mental tasks are unfortunately seldom combined in brain imaging.

Existing neuroimaging methods typically perform either discovery of unknown neural structure or testing of neural structure associated with mental tasks.

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 autoencoder. 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.