Features

AGNOSTIC LEARNING vs. PRIOR KNOWLEDGE
http://www.agnostic.inf.ethz.ch/

"When everything fails, ask for additional domain knowledge"

Does adding domain/prior knowledge really help or can we use as well "dumb" features and a good learning algorithm?

You will get to compete on 5 classification problems for which you are provided with both the raw data and preprocessed data. Some of the raw data does not come in a feature representation (e.g. chemical molecules are described as the position of their atoms in space). You get information about the data so you can extract smart features and boost your classifiers performance with domain/prior knowledge. Alternatively, you can rely on the preprocessed data and work on improving your generic classification algorithms.

Workshops
The results of the challenge will be discussed at 2 workshops:

1) NIPS 2006 workshop on multi-level inference
http://clopinet.com/isabelle/Projects/NIPS2006/

The agnostic learning track of the challenge is used to implement a model selection game. Win one of two prizes (and get to ski at Whistler!)

2) IJCNN 2007 workshop on agnostic learning vs. prior knowledge
http://clopinet.com/isabelle/Projects/agnostic/ Other prizes will be awarded (TBA).

Proceedings

To submit a full length paper on model selection, see the call for paper:
http://clopinet.com/isabelle/Projects/modelselect/call-for-papers.html

The results of the challenge can be submitted to IJCNN 2007 to be published in the proceedings

The book on feature extraction with the results of the first challenge is now available:
http://clopinet.com/fextract-book/

We are looking forward to your participation!

The organizers