Detecting psychiatric disorders with statistical learning tailored to brain activity

Gael Varoquaux
Institut National de Recherche en Informatique et Automatique (INRIA)

Psychiatric disorders have no clear boundaries or visible impact on the tissues. There are on-going efforts to define them by comparing the brain activity of individuals across a population. For psychological standpoint, such effort requires linking individual traits to data on their brain activity. Brain activity at rest in particularly promising, as it can easily be accumulated and compared across subjects.
I will cover machine-learning, signal-processing, and algorithmic aspects of this research agenda. Machine learning will touch upon simple schemes for predicting multiple traits jointly, ie multi-task learning; signal processing will cover modeling the variation of covariance matrices, using information geometry, and algorithmic considerations will touch upon processing efficiently the resulting multi-terabytes sets of data, using stochastic optimization.

Presentation (PDF File)

Back to Workshop I: Big Data Meets Large-Scale Computing