Databases Reference
In-Depth Information
A distributed system offers resource-sharing capabilities and is able to adapt
to the incremental growth of resources. It also provides reliable resource avail-
ability. Distributed systems have evolved from topological networks, such as
the Ethernet, to the current cloud computing technology. Cloud computing
is a form of utility computing that enables large-scale activities and opera-
tions to be performed beyond the boundaries of local networks or within a
particular organizational structure. The large amount of resources available
within the computational cloud provides the means for performing complex
large-scale engineering or scientific analyses, including recognizing data at the
Internet-scale.
Parallel computing technology addresses how information can be processed
more e
ciently. Our universe can be viewed as a parallel computer, in which
all of the elements perform their operations simultaneously. In the last sev-
eral decades, parallel computing technology has advanced from simple multi-
threading computations to multi-core and graphical processing unit (GPU)
computing technologies. By performing concurrent computations, parallel
computing enables large-scale problems to be solved more e
ciently. For ex-
tensive pattern recognition processes on existing Internet-scale data, parallel
computing is a similar to the discovery of fire.
The rapid development of machine intelligence schemes is also contribut-
ing to the development of Internet-scale pattern recognition. From classical
statistical approaches to sophisticated machine learning techniques, compu-
tational intelligence schemes are critically important in numerical analysis
and other experimentations involving scientific data. Nevertheless, a key issue
that needs to be addressed is the scalability of such schemes when processing
Internet-scale data.
This topic describes fundamental research on the scalability of pattern
recognition. Scalability in the context of pattern recognition can be defined
as the ability to either handle growing amounts of patterns in a graceful man-
ner or to be readily enlarged. The scalability issues of the existing pattern
recognition schemes for Internet-scale data deployment will be presented. A
number of different approaches will be extensively reviewed, and a number of
possible solutions for the scalability problem will be introduced.
1.3 Computational Intelligence Approach for Pattern
Recognition
The development of computational intelligence schemes can be traced back
to the first computational intelligence test conducted by Alan Turing in 1950.
Computational intelligence involves schemes that use a set of procedures and
operations to mimic the intelligence of biological organisms. One of the best
