Picon

GSoC Introduction - Generic SO Framework

Hi all,


My name is Fernando Iglesias and I am a CS/EE student from Spain living in Sweden.
You maybe recognize my nickname in IRC, n4nd0. As you may know, I have been 
selected for the awesome task of developing a generic framework for Structured Output
learning into Shogun mentorized by Nico Goernitz.

A short summary of the project divided into its main steps is, according to my view:
  • Implement generic SO-SVM as it is presented in Tsochantaridis et. al. paper. My
    idea here is to use libqp whose functionality have been recently updated in Shogun.
  • Two cases of use for SO-SVM, multiclass SVM and HM-SVM. 
  • Develop a second method for SO learning, Conditional Random Fields.
For some time now, I have been studying the theory of SO learning view and now I plan
to start an "on paper" design of how the framework should fit into Shogun. Any thoughts
and suggestions in this or other aspects are very welcome.

In addition to contact with the mentor and other Shogun contributors through mailing list
and the IRC channel, I plan a closer collaboration with Viktor Gal and Michal Uricar, also
GSoC students whose projects are somewhat related to generic SO.

I am really happy to know that Gunnar Raetsch, one of the authors of an implementation
of HM-SVM, is one of Shogun's authors. I hope to learn a lot from him and the other ML 
experts within Shogun.

Thank you very much for this wonderful project.

All the best,
Fernando.

Gmane