CS 68191 Masters Seminar / CS 89191 Doctoral Seminar
Spring 2007
Masters Student Presentation
Trend Motif Presentation Abstract
Scott McCallen
Systems of interacting units across many fields of science can be
described as complex networks. These complex networks are often so
large that it is hard to immediately understand how the individual
units cooperate as a whole. One method to understand the underlying
structure is to identify the reoccurring subgraphs in the large
network. These network motifs have been found in many fields of
science, including biochemistry, neurobiology, ecology, and
engineering [1], and they have been demonstrated to be very likely to
form the basis of the entire network. However, many large networks in
the real world are not static, but exhibit dynamic properties. In such
a setting, weights are often associated with the vertices or edges of
the network, and change constantly. Formally, the weight for each
vertex or edge forms a time series. In this situation, the focus of
motif search should consider not only network topology, but also the
dynamics of its weight information. Indeed, how to capture the
dynamics of the network through motifs is an open problem.
In this work, we propose a formal definition for trend motif which
takes both network topology and time series information into
consideration. Specifically, the trend motif describes reoccurring
subgraphs, where each of its vertices or edges show similar dynamics
in a user-defined period. We further classify the trend motifs into
several categories, and provide a uniform framework to enable users to
query these different types of motifs. The efficient algorithm to
search for these motifs in very large dynamic network is currently
being developed.
References
[1] S. Shen-Orr, R. Milo, S. Mangan, and U. Alon. Network motifs:
Simple building blocks of complex networks. Science, (298):824–827,
2002.