Databases Reference
In-Depth Information
fore, reducing the complexity of the learning mechanism is an objective
of scalability. Examples of improving scalability using the learning ap-
proach include active learning [16] and incremental learning [11]. A sig-
nificant limitation of this approach is that the accuracy of the algorithm
may be sacrificed for the sake of fast and simple learning capabilities.
3. Distributed Computing Approach: The advancement in networking tech-
nologies has enabled large-scale computations to be performed within
the body of a network itself. Rapid developments in high performance
computing and grid technologies allow a collaboration of resources to
work for a specific application. In this context, existing pattern recogni-
tion algorithms may be implemented on a distributed computing plat-
form using parallel processing. Some examples of scalable pattern recog-
nition schemes using this approach include the works carried out in
[1, 17].
1.4.3 Distributed Computing Solution for Scalability of PR
Schemes
The distributed computing approach for scaling existing pattern recognition
algorithms has the potential to be the optimum solution. However, some of
the existing algorithms are highly complex and di
cult to parallelize. Devel-
opments of neural network algorithms for pattern recognition have provided
an interesting insight into the implementation of pattern recognition in dis-
tributed computing. In their nature, neural networks are formed through the
collaboration of computational nodes, known as neurons. Due to the tightly
coupled nature of existing neural network schemes, the integration of these two
components is still in its infancy. It was initially conceived for single-processing
(CPU-centric) architectures, which rely heavily on iterative techniques. There-
fore, more work is needed to attain the effectiveness and e
ciency of neural
network algorithms for pattern recognition using the distributed computing
approach.
Given the rapid advancement in existing distributed processing technolo-
gies, distributed computing may provide seemingly unlimited scalability for
large-scale processing. Implementations of pattern recognition schemes in a
distributed manner are possible in a variety of distributed computing environ-
ments using a simple, computationally inexpensive, and embarrassingly par-
allel recognition algorithm. Therefore, distributed pattern recognition (DPR)
may be a solution for Internet-scale pattern recognition. Further discussions
on the distributed computing approach will be presented in Chapter 2.
