Date: Fri, August 3, 16:00-18:00 Place: Room Ew501, E Block, IIS, The University of Tokyo Invited Speaker: Prof. Maciej Ogorzalek (Jagiellonian University) Title: Ensambling techniques for time-series-base model building (with bio-medical applications) Abstract: This talk shows the application of advanced Statistical Learning methods to the fields of Bio- and Cheminformatics. After reviewing the concept of Statistical Learning, several strategies for minimizing the generalization error are discussed, focusing on ensemble methods, where the outputs of several models are combined to an averaged output. The Matlab toolbox ENTOOL for ensemble modeling is introduced which allows to quickly access a broad range of Statistical Learning techniques under a unified interface. Several application examples are presented where ENTOOL is used . for detecting relevant sequence positions . for building nonlinear models of skin permeability Finally the Molecular Graph Network (MGN) for direct mapping from molecular graph structure to a desired output property is presented that constructs adaptive descriptors and can be used to for classification and regression on chemical compounds.