Home
Reading
Searching
Subscribe
Sponsors
Statistics
Posting
Contact
Spam
Lists
Links
About
Hosting
Filtering
Features Download
Marketing
Archives
FAQ
Blog
 
Gmane
From: Viktor Gal <wiking-xmXv9IxLUjkAvxtiuMwx3w <at> public.gmane.org>
Subject: GSoC Introduction - Latent SVM
Newsgroups: gmane.comp.ai.machine-learning.shogun
Date: Tuesday 24th April 2012 14:18:40 UTC (over 4 years ago)
Heeey,

::: i'm Viktor (you might know me by the nick wiking on IRC). I'm currently
a PhD student working on image classification problems, mainly focusing on
medical images. I've been selected to implement a general purpose latent
SVM as part of the GSoC of shogun. My mentor for the project is Alexander
Binder. Some of you might know that for the research in SVM with latent
variables Pedro Felzenszwalb won the PASCAL VOC "Lifetime Achievement"
Prize:
http://pascallin.ecs.soton.ac.uk/challenges/VOC/voc2010/workshop/

My project basically consist of 2 bigger parts:
 1) implement a basic latent svm solver: this part is actually almost done,
just some code cleanup is required.
 2) extending the basic implementation to allow: 
      . .. structured output: for this i'll work together with Fernando,
who's doing the SO framework GSoC project of shogun. compare it with
current implementations [1]
      . .. Max-Margin Min-Entropy Models as proposed in [2]
      . .. implement static interfaces for important application cases

all the code that's gonna be produced during this project is and will be
available at: 
https://github.com/vigsterkr/shogun/tree/latent

::: as currently i'm mainly working with image classification problems, the
example applications of the module is going to be from this field. Thus, if
you would like to apply it to problems from any other fields (e.g. text
processing), please contact me and i'll be more than happy to help as well
as include it in examples code!

::: apart from this and still part of the GSoC project i'll try to do a
code refactoring, in collaboration with Fernando and Michal, for a better
integration of various QP solvers, focusing on libqp.

::: at last but not least, thanks a lot for the confidence and support for
giving me the opportunity to work on this project during this summer! i
hope to see a lot of discussions about the future and further improvement
possibilities of shogun on IRC like in the last 1-2 weeks! 

cheers,
viktor

[1] C.-N. Yu and T. Joachims, Learning Structural SVMs with Latent
Variables, International Conference on Machine Learning (ICML), 2009
[2] Kevin Miller, M. Pawan Kumar, Ben Packer, Danny Goodman, Daphne Koller,
Max-Margin Min-Entropy Models, JMLR W&CP 22: 779-787, 2012
 
CD: 3ms