Join us for an invited talk with Jussi Tohka (CONEX professor/distinguished researcher; Bioengineering and Aerospace Engineering Department; UC3M):
- Title: “Machine learning in brain imaging”.
- Event Date: Monday, June 15; 2015, 12:00 am
- Location: 4.2.E03 Room; Torres Quevedo Building; Leganés Campus; Universidad Carlos III de Madrid.
In brain imaging supervised learning algorithms are essential in trying to understand complex cognitive processes, developing imaging based biomarkers for brain diseases, and in designing brain computer interfaces. The development of new machine learning approaches is partly driven by new massive, openly available brain imaging databases that are revolutionizing the field and creating new opportunities for method developers. The data are noisy, heterogeneous, and high-dimensional, i.e., the number of variables exceeds the number of samples, making the learning problems ill-posed. The most efficient use of this new data requires the development of new learning algorithms that are capable to utilize imaging specific properties of the data while solving problems related to high dimensionality and heterogeneity of the data and combine non-imaging (psychological test results, genetic) information to brain imaging information.
In this talk, I will give an overview of potential applications of machine learning techniques in brain imaging and provide ideas how to solve the variable selection problem while taking the spatial structure of the data into account. I will also highlight the high variance of the standard cross-validation based model selection approach and propose to use a new parametric Bayesian error estimator instead. Finally, I will describe the problems arising when using machine learning methods for disease classification in aging or developing populations.