Databases Reference
In-Depth Information
Chapter 8
Pattern Recognition within
Coarse-Grained Networks
A distinctive difference between conventional and distributed pattern recog-
nition is the resource consideration. In a distributed approach, the system
must be capable of utilizing the available resources effectively and e
ciently.
A good communication model must be considered to ensure proper utiliza-
tion and communication of resources between processing nodes. Distributed
pattern recognition (DPR) has the ability to scale up the process as the size
of the problem increases. However, the scalability depends on the resource
availability in a particular computational network. Resource availability is
influenced by the capacity and stability of the computational network. The
network capacity in distributed applications, such as DPR, is observed in
terms of the granularity of the network. Commonly, computational networks
are either coarse-grained, such as a computational grid, or fine-grained, such
as a wireless sensor network (WSN). The processing capacities and capabili-
ties of these networks differ. Because the application deployment focused on
a single problem domain, most existing pattern recognition schemes are non-
adaptive to different network granularities. The DHGN pattern recognition
scheme described in Chapter 5 was developed with adaptive network granu-
larity consideration [4] and the algorithm can be deployed in both coarse- and
fine-grained networks.
In this chapter, we will look at the network granularity aspect of distributed
pattern recognition (DPR). We will demonstrate how the DHGN algorithm
can be deployed in a network of different granularity, which allows for flexible
recognition of different forms of Internet-scale data. In addition, we will discuss
specific pattern recognition applications in coarse-grained networks.
8.1 Network Granularity Considerations
Granularity of a computational network refers to the levels of its compo-
sition. A coarse-grained network mainly consists of a few large processing
entities, which are capable of handling significantly high computational loads.
An example of this type of network is a computational grid network. Con-
121
