2-go grudnia br. (piątek) o godz. 12:00 w Auli COK Wydziału Fizyki, Astronomii i Informatyki Stosowanej UMK odbędzie się wykład pt. Online Learning with Multiple Kernels, który wygłosi dr Masahiro Yukawa z Uniwersytetu Keio w Tokio i Uniwersytetu Technicznego w Berlinie.
I will present a brief introduction to the multikernel adaptive filtering framework that has been proposed for the adaptive estimation task of nonlinear functions of Euclidean vectors. A major difference from the multiple kernel learning (MKL), besides its adaptive aspects, is its higher degrees of freedom, which is r (expansion length) times Q (# kernels) rather than r + Q. This allows us to seek for the best point, in the mean squared error sense, in the sum space of the subspaces spanned by the Q sets of the basis vectors that are associated with the Q kernels, respectively. I will also talk about the regularization that is clearly important when many kernels are employed.