2pm, Tuesday, 25 Apr 2006, in Meyer 1024:

<CENTER><B>
Supervised Ranking With L_p Norms
</B>
<P>
Cynthia Rudin
<BR>
LCV

</CENTER>


The problem of supervised ranking is useful in many application 
domains, e.g., natural language processing, customer service routing, 
and bioinformatics. Many of these domains require the construction of 
a ranked list, yet often, only the top portion of the list is used in 
practice. In this work, our goal is to design supervised ranking 
algorithms that perform especially well near the top of the ranked 
list.
<P>
First, I'll provide a general form of convex objective that gives high-
scoring examples more importance. This "push" near the top of the list 
can be chosen to be arbitrarily large or small; L_p-norms are used to 
provide a specific type of push. I will then derive a corresponding 
boosting-style algorithm (the "P-Norm Push Algorithm"), and illustrate 
the usefulness of the algorithm through experiments on publicly 
available repository data. Finally, I will present a generalization 
bound based on the p-norm objective (with a brief proof outline), and 
a theorem stating that the algorithm's objective is unique in a 
specific sense.

<P>
<I><A HREF="index.html">Back to LCV meeting schedule</A></I>