Pal S.K., Ray S.S., Ganivada A. Granular Neural Networks, Pattern Recognition and Bioinformatics

Springer, 2017. — 235 p.The volume “Granular Neural Networks, Pattern Recognition and Bioinformatics” is an outcome of the granular computing research initiated in 2005 at the Center for Soft Computing Research: A National Facility, Indian Statistical Institute (ISI), Kolkata. The center was established in 2005 by the Department of Science and Technology, Govt. of India under its prestigious IRHPA (Intensification of Research in High Priority Area) program. Now it is an Affiliated Institute of ISI.
Granulation is a process like self-production, self-organization, functioning of brain, Darwinian evolution, group behavior and morphogenesis—which are abstracted from natural phenomena. Accordingly, it has become a component of natural computing. Granulation is inherent in human thinking and reasoning process, and plays an essential role in human cognition. Granular computing (GrC) is a problem-solving paradigm dealing with the basic elements, called granules. A granule may be defined as the clump of indistinguishable elements that are drawn together, for example, by indiscernibility, similarly, proximity or functionality. Granules with different levels of granularity, as determined by its size and shape, may represent a system differently. Since in GrC, computations are performed on granules, rather than on individual data points, computation time is greatly reduced. This made GrC a very useful framework for designing scalable pattern recognition and data mining algorithms for handling large data sets.
The theory of rough sets that deals with a set (concept) defined over a granulated domain provides an effective tool for extracting knowledge from databases. Two of the important characteristics of this theory that drew the attention of researchers in pattern recognition and decision science are its capability of uncertainty handling and granular computing. While the concept of granular computing is inherent in this theory where the granules are defined by equivalence relations, uncertainty arising from the indiscernibility in the universe of discourse can be handled using the concept of lower and upper approximations of the set. Lower and upper approximate regions respectively denote the granules which definitely, and definitely and possibly belong to the set. In real-life problems the set and granules, either or both, could be fuzzy; thereby resulting in fuzzy-lower and fuzzy-upper approximate regions, characterized by membership functions.
Granular neural networks described in the present book are pivoted on the characteristics of lower approximate regions of classes demonstrating its significance. The basic principle of design is—detect lower approximations of classes (regions where the class belonging of samples is certain); find class information granules, called knowledge; form basic networks based on those information, i.e., by knowledge encoding; and then grow the network with samples belonging to upper approximate regions (i.e., samples of possible as well as definite belonging). Information granules considered are fuzzy to deal with real-life problems. The class boundaries generated in this way provide optimum error rate. The networks thus developed are capable of efficient and speedy learning with enhanced performance. These systems have a strong promise to Big data analysis.
The volume, consisting of seven chapters, provides a treatise in a unified framework in this regard, and describes how fuzzy rough granular neural network technologies can be judiciously formulated and used in building efficient pattern recognition and mining models. Formation of granules in the notion of both fuzzy and rough sets is stated. Judicious integration in forming fuzzy-rough information granules based on lower approximate regions enables the network in determining the exactness in class shape as well as handling the uncertainties arising from overlapping regions. Layered network and self-organizing map are considered as basic networks.Introduction to Granular Computing, Pattern Recognition and Data Mining
Classification Using Fuzzy Rough Granular Neural Networks
Clustering Using Fuzzy Rough Granular Self-organizing Map
Fuzzy Rough Granular Neural Network and Unsupervised Feature Selection
Granular Neighborhood Function for Self-organizing Map: Clustering and Gene Selection
Gene Function Analysis
RNA Secondary Structure Prediction: Soft Computing Perspective