NOTICE:

The current code is not completely ready for release. 
In particular, multiple instance learning is not implemented in this version 
because the code used for the article is not clean yet.

As a consequence, this version may contain bugs. Please email the author (armand joulin at firstname.lastname@ens.fr) 
if something does not work.

REQUIREMENT:

-toolboxes:

The code requires Mark Schmidt's toolbox for the rounding (EM procedure):

http://www.di.ens.fr/~mschmidt/

Note that in its final version, this code will also require Julien Mairal's toolbox (for multiple instance learning).

-data:

For semi supervised learning (SSL), download files at:

http://olivier.chapelle.cc/ssl-book/benchmarks.html

and put them in the input/SSL directory.

For multiple instance learning, the link will be given when the code is added.

FILE TO LAUNCH:

main_clustering : An example of clustering with our method
noise_clustering.m : Comparison with k-mean for clustering with an increasing level of noise
semi_sup.m : An example of semi supervised learning

CITATIONS:

If you use our code:

@InProceedings{JouBacICML12,
   title = "A convex relaxation for weakly supervised classifiers",
   booktitle = "Proceedings of the International Conference on Machine Learning (ICML)",
   author = "A. Joulin and F. Bach",
   year = "2012"
}

If you use the SSL data please cite the excellent book by Chapelle et al.:

@Book{ChaSchZie06,
  editor =   {O. Chapelle and B. Sch{\"o}lkopf and A. Zien},
  title =    {Semi-Supervised Learning},
  publisher =   {MIT Press},
  year =     2006,
  url =          {http://www.kyb.tuebingen.mpg.de/ssl-book},
  address =  {Cambridge, MA}
}


