Databases Reference
In-Depth Information
2. Over-fitting problem: A small training set cannot represent actual large-
scale data.
3. The recognition function within neural networks or machine learning
works only for a specific problem within the recognition domain. Net-
work retraining is required for different sets of problems.
The existing literature on computational intelligence shows that the scalabil-
ity and adaptability of machine learning approaches outweigh other pattern
recognition approaches. Some machine learning approaches can be used in
parallel environments, and recognition loads are distributed across computa-
tional nodes. Nevertheless, these algorithms require extensive and complex
computations to derive the best solution for the recognition process.
1.4 Scalability in Pattern Recognition
The emergence of the data deluge phenomenon has brought forward the
need for recognition schemes for Internet-scale patterns. We can divide large-
scale pattern recognition into two perspectives. The first perspective is recog-
nition of a large number of patterns. In this context, the focus is on the volume
of the patterns. The recognition process recognizes or classifies patterns into
a large number of clusters, and a large number of patterns are stored. The
second perspective is recognition of large patterns. In this case, the pattern
data are huge, as in the areas of face or image recognition.
In this section, we consider some of the common barriers encountered when
implementing pattern recognition for large-scale data sets and some of the
possible solutions. Our interest lies in the implementation of widely avail-
able distributed computing infrastructures for scalable pattern recognition
schemes.
1.4.1 Common Barriers
Pattern recognition is an important tool in evaluating and analyzing large-
scale data that have been produced in a wide range of applications. Neverthe-
less, current approaches incur excessive computational complexity to adapt to
these large and highly complex data sets. When processing large and highly
complex data sets, there are a number of barriers that need to be addressed
with respect to the implementation of pattern recognition. These include the
following:
1. Size of data: As the size of the data set increases, existing pattern recog-
nition schemes must be able to manage data in an e
cient manner with
specific concerns for storage and transport. Methods in a recognition
