Hello,
Are you good at hierarchical clustering/causal discovery/graphical
modeling/link analysis/phylogenetics/string-matching? Here is your
chance to prove it!
*The Pascal Challenge on Computer-Assisted Stemmatology* evaluates
methods for reconstructing the family-tree of a group of related
documents. Such a family-tree corresponds to a) a clustering hierarchy,
where joined subgroups make subtrees; b) a causal/graphical model of
interdocument dependencies; c) a network of information flow among the
documents; d) a phylogenetic tree; etc. Many of the applicable
techniques are often applied in unsupervised scenarios. The Challenge
presents an opportunity to compare different methods and approaches in a
supervised and objective fashion (and to show that your approach is the
best).
A prototypical example illustrating the problem is as follows: A top AI
researcher has finally concluded (after, what, 20 years?) that the
statement "Tweety is a bird" is true. The rumour of this fact spreads
around like wildfire, becoming distorted along the way. After a while, a
set of scientists submit papers to UAI'08 claiming respectively that:
"Sweety is a bird"; "Sweety has a bird"; and "Tweety is a nerd". Can we
deduce how the information spread among the scientists? (No, we are not
interested to know if Tweety can fly... Sorry.)
Participation is open to all.
More information can be found on the web-page of the Challenge:
http://www.cs.helsinki.fi/teemu.roos/casc/
The schedule is as follows:
First-phase data available October 6, 2006
Second-phase data available November 30, 2006
Validation data available * February 20, 2007 *
Submission deadline March 30, 2007
Results April 30, 2007
So as you see, the data is already available, including a validata set
and a correct (and an incorrect) solution for it:
http://www.cs.helsinki.fi/teemu.roos/casc/example.html
We invite applications of established and, in particular, novel
approaches to stemmatology, including but of course not restricted to
hierarchical clustering, graphical modeling, link analysis,
phylogenetics, string-matching, etc.
Quoting Prof. Michele Sebag (manager of the Pascal Challenges
Programme): ''Beside the fact that Pascal Challenges are widely known
and enjoyable, they also provide the ML community with new and smart
benchmarks.''
We hope to live up to these expectations.
Organizers:
* Teemu Roos, Helsinki Inst. for Information Technology
* Tuomas Heikkilä, Dept. of History, University of Helsinki
* Petri Myllymäki, Dept. of CS, University of Helsinki
On behalf of the organizers,
Teemu Roos
Helsinki Institute for Information Technology
_______________________________________________
uai mailing list
[email protected]
https://secure.engr.oregonstate.edu/mailman/listinfo/uai
