Statistische und Probabilistische Methoden der Modellwahl

Abstract

Aim of this conference with more than 50 participants, was to bring together leading researchers from roughly three different scientific communities who work on the same issue, data based model selection. Their different methodological approaches can be roughly classified into (1) Frequentist model selection and testing (2) Statistical learning theory and machine learning (3) Bayesian model selection The key task in model selection is to select a proper mathematical model based on information generated by data and/or by prior knowledge. Proper might mean a model with minimal prediction error, a model which describes the main qualitative data features, such as bumps and modes, or a model of low computational complexity. Mathematical techniques and concepts encountered with this workshop are wide spread, ranging from concentration and oracle inequalities, asymptotic analysis and distribution theory to testing theory, information measures and nonconvex optimization.