Neuroimaging techniques produce large amounts of brain images of different natures, allowing researchers and clinicians to gain insights of unprecedented quality on the cerebral anatomy, its connectivity structure and its functions. On the one hand, the development of these techniques provides neuroscientists with a growing amount and variety of data, and thus, a potentially improved understanding of the brain, and on the other hand, it precisely poses the challenge of devising automated methods for a high-level understanding of neuroimages. These methods would be of importance to decode mental thoughts, understand cortical representations, categorize and classify brain responses, detect abnormalities in the brain, remove noise, take advantage of correlated prior information, help the diagnosis, and so on. Machine learning is probably one of the most promising field of research that would bring new approaches and procedures for automated neuroimaging interpretation.

The main goal of this workshop is precisely to bring together people from the machine learning community and people from the neuroimaging community that are keen to discuss their expertises. Potential outcomes to this workshop are for instance: the formal/machine learning setting of common problems in neuroimaging, the identification of new problems that can be readily tackled using machine learning techniques, the creation of new collaborations. It is also expected that discussions will build around important challenges of machine learning posed by neuroimaging data such as feature selection in presence of few data, transfer learning, structured prediction...

Among the various themes that are of primary interest for the workshop, time will be devoted to sparsity based methods, feature selection, graph-based representation of image and kernel methods, exploitation of prior and heterogeneous knowledge to build predictive models.

Liva Ralaivola (LIF, Marseille, France)

Bertrand Thirion (Parietal / Inria, Gif sur Yvette, France)