![](bg1.png)
REAL TIME CLUSTERING MODEL
J. Cheng, M. R. Sayeh
Department of Electrical and Computer Engineering
Southern Illinois University Carbondale, Carbondale, IL 62901, U.S.A.
M. R. Zargham
Department of Computer Science, Southern Illinois University Carbondale, Carbondale, IL 62901, U.S.A.
Keywords: ODE, Clustering, Vector quantization, Real time.
Abstract: This paper focuses on the development of a dynamic system model in unsupervised learning environment.
This adaptive dynamic system consists of a set of energy functions which create valleys for representing
clusters. Each valley represents a cluster of similar input patterns. The system includes a dynamic parameter
for the clustering vigilance so that the cluster size or the quantizing resolution can be adaptive to the density
of the input patterns. It also includes a factor for invoking competitive exclusion among the valleys; forcing
only one label to be assigned to each cluster. Through several examples of different pattern clusters, it is
shown that the model can successfully cluster these types of input patterns and form different sizes of
clusters according to the size of the input patterns.
1 INTRODUCTION
As stated in (Jain, 1988), "Cluster analysis is the
process of classifying objects into subsets that have
meaning in the context of a particular problem." In
other words, clustering is a process of grouping a set
of unlabeled data. As shown in Figure 1, in general,
clustering can be grouped into two types: non-
overlapping (exclusive) and overlapping
(nonexclusive). In non-overlaping, each object input
will be assigned to only one cluster whereas in
overlapping an object can be assigned to more than
one cluster. In this paper we only consider non-
overlapping clustering. Non-overlapping clustering
could lie either intrinsic or extrinsic. In the intrinsic
approach, also called unsupervised learning, a
proximity matrix is the only criteria used. (Proximity
matrix represents relationship between the objects; if
the objects are patterns such matrix could represent
the distance between the patterns). The extrinsic
approach, also called supervised learning, in
addition to proximity matrix, it also uses category
labels on the objects. To notice the difference
between these two approaches, let’s consider a set of
data representing
health condition of normal and
overweight children. Using intrinsic approach, we
can group these children based on these factors and
then try to determine whether overweight plays a
role in academic status. Taking an extrinsic
approach allow us to study the way of separating
normal and overweight children by considering their
health conditions.
We consider intrinsic approaches only. The
intrinsic methods can be
divided into five types:
hierarchical, partitional, grid-based,
artificial neural
networks, and evolutionary (Jain, 1988; MacQueen,
1967; Grossberg, 1976; and Kohonen, 1982).
Comparison of these clustering methods is hard
to do using simulation because of different
implementation of the methods and the data that is
used. It is also hard to do theoretically comparison
of them because they are almost impossible to model
mathematically (Jain, 1988). Furthermore, the
existing models impose architectural complexity
and/or time complexity which prevent them of
having real time response time. To overcome the
real time mathematical modeling problems, we have
proposed a new method which depends solely on
ordinary differential equations (ODE) (Cheng, 2006).
There is no need for IF/THEN logical statements.
Therefore it can be easily implemented on the
analog type devices to take advantage of high-speed
electronics or photonics technologies.
235
Cheng J., R. Sayeh M. and R. Zargham M. (2008).
REAL TIME CLUSTERING MODEL.
In Proceedings of the Tenth Inter national Conference on Enterprise Information Systems - AIDSS, pages 235-240
DOI: 10.5220/0001694002350240
Copyright
c
SciTePress