Kernel methods in dynamic classification problems

Contact person: Khalid Daoudi
Synopsis:

• Kernel methods have shown their power in static classification problems. The best example is definitely the support vectors machine (SVM) which is the most popular algorithm for supervised classification of static data. In many applications however (such as speech, video, bioinformatics...), one needs to classify dynamic sequences of observations. Classical kernel methods fail to provide satisfactory solutions in such applications. The gaol of this e-tam is to develop new kernel-based methods for the classification of dynamic sequences of observations. The ultimate (and ideal) goal being to develop a general theory/formalism of sequence kernels that leads to powerful dynamic classification algorithms.
  

Keywords:

• Kernel methods, sequence kernels, string kernels, kernels between sets, SVM, supervised learning, classification.
  

Additonal info: E-Team's webpage