Learning when the data are functions : operator-valued kernels, function-valued RKHS, and beyond.
Résumé : In this talk, I will discuss concepts and methods of kernel-based learning for functional data. The focus is on the case where covariates as well as responses are functions. Basic concepts of RKHS theory are extended to the domain of functional data analysis and the conditions under which such an extension is feasible are discussed. Our main results demonstrate how basic properties of kernel-based classification and regression known from multivariate statistical analysis can be restated for functional data, if appropriate conditions are satisfied.
Travail joint avec E. Duflos (LAGIS-EC Lille/CNRS), P. Preux (SequeL-INRIA Lille), S. Canu (LITIS-INSA Rouen)
Cet exposé se tiendra en salle C20-13, 20ème étage, Université Paris 1, Centre Pierre Mendès-France, 90 rue de Tolbiac, 75013 Paris (métro : Olympiades).