Seminar of Institute of Information Systems and Applications

Speaker: Prof. Hiroshi Mamitsuka,
Bioinformatics Center,
Institute for Chemical Research Kyoto University

Topic: Clustering over graphs: Probabilistic model based approach

Date: 13:30 to 15:00. Wednesday, 23-Nov-2011

Place: 103 Delta Building (台達館)

Abstract:
Graphs are recently emerging and currently abundant data,
including websphere, blogosphere, social networks and gene networks
and so on. In this talk, I will address a problem of clustering nodes
given multiple graphs, which share the same node set and
gold-standard clusters. One practical feature on clustering with
multiple graphs is gold standard clusters are likely to appear only
part of given multiple graphs. We call these clusters local clusters.
For solving this clustering problem with local clusters, I will pursue
an approach of using probabilistic model learning by which local
clusters can be captured and preserved. My talk will start with a
review on probabilistic model-based approaches for clustering nodes
over a single graph. I will then move on to the main subject of this
talk, i.e. clustering nodes from multiple graphs. I will show a
problem setting we consider and present a new approach, based on
probabilistic model learning, which allows to clearly outperform
possible approaches under a lot of experimental settings with both
synthetic graphs and real gene networks.

All faculty and students are welcome to attend.