Speaker: Hiroshi Mamitsuka, Kyoto University
Time: 1:30pm-2:30pm, Friday, August 10th, 2007
Place: 321 Stanley Hall
Abstract:
In this talk, I address the issue of clustering numerical vectors with
a network. This is a very popular problem setting, and our focus is on
the optimal combination of two heterogeneous data sources. An example
of this setting is clustering genes whose behavior can be numerically
measured and a gene network can be given from another data source. Web
pages can be also numerically vectorized by their contents, e.g. term
frequencies, and at the same time, are hyperlinked to each other,
showing a network. I'll talk two different approaches, in each of
which a new graph clustering measure is defined based on the network
modularity, a recently proposed concept in the field of graph
theory. The first approach is based on a probabilistic model which was
extended from a hidden Markov random field, and the second is based on
the idea of spectral clustering on graphs. The performance of the
proposed methods was measured by empirical experiments using
synthetic as well as real-world data from molecular biology.
Experimental results showed that the methods are effective enough to
have good results for the issue addressed. The extended abstracts of
this talk is to appear in ISMB and KDD this year.
To receive announcements of future seminars in this series, send a message to
majordomo@listlink.berkeley.edu with the following in the body of the
message:
subscribe bioinformatics your_email_address
There should be no other text or signature files in the body of the message.