SV Kernels from Hidden Markov Models and Probabilistic Context-Free Grammars
Dr Chris Watkins, Department of Computer Science, Royal Holloway, University of London

Abstract: Kernel functions enable linear approximation methods to be used for apparently non-linear problems. I will explain what kernel functions are and why they can be useful. I will then describe a new family of kernel functions based on pair HMMs and pair PCFGs. These kernels can potentially allow linear statistics to be used in a natural way to learn classifiers and regression functions for bio-sequences, speech, and sentences.

This seminar was held at the Department of Computer Science, Royal Holloway, University of London on 17 December 1998.

back