Research Seminar
Sparse simultaneous component analysisRobert van den BergKU Leuven | |
| Abstract: | In contemporary science, there is a trend to collect more and more complex data. For instance, within psychology one may collect for the same participants blocks of questionnaire, fMRI, and physiological data. This may lead to hundreds of variables per participant. These data then may be subjected to analyses to identify the (psychological) mechanisms underlying the different blocks of measurements. Methods of simultaneous component analysis (SCA) may be suited for this purpose, as they extract the most significant patterns from multiblock data by simultaneously analyzing all measurement blocks. In SCA methods, all variables are allowed to contribute to the underlying components. However, it is not likely that this is the case, indeed; moreover, if all variables would contribute to all components nonetheless, in view of the hugue number of variables, it would be most tedious to study the full set of variable contributions. To address this problem, we introduce sparse SCA. Sparse SCA is a generalization to the family of SCA methods of sparseness techniques that were previously introduced in regression and principal component analysis. In this presentation we will outline the principles of sparse SCA, the associated algorithm, an approach to model selection, and some results from a systems biology case study. |
| Date: | Tue Mar 2, 12:15 pm - 1:15 pm |
| Place: | room 02.51 (Department of Psychology, Tiensestraat 102, 3000 Leuven) |