The objective of this chapter is the introduction of artificial neural networks in the context of directed graphs. In particular, a linkage from graph theory through signal flow graphs to artificial neural networks is provided. Within the context of pattern recognition, a number of feed-forward neural based approaches are introduced and discussed. Motivation leading to the design of each neural method is also given. The main contribution of this book chapter is the provision of a basic introductory text with less mathematical rigor for the benefit of students, tutors, lecturers, researchers, and/or professionals who wish to delve into the foundational representations, concepts, and theory of bio-inspired intelligent systems.
TopIntroduction
Unconsciously, humans and computer systems generate massive amounts of data each day. From airline reservation systems through e-Commerce portals to genome research projects, varying degrees of data are generated each day. The bulk being amorphous in nature, exist in the form of emails, text files, video, speech etc. With researcher estimate of institutions amorphous data close to 80 percent, it is therefore prudent to mine such data kinds to support business decisions. The inception of the World Wide Web (WWW) has also seen a drastic increase in amorphous data production. Today, the Internet is considered the largest data repository in the world. News giants such as BBC, CNN, etc publish news evolving around the world on the internet, in a relatively short time. Breaking news which in the past, took days to circulate, can now be disseminated in a matter of minutes. Social networks such as facebook, twitter, etc. have served the purpose of information dissemination for most individuals, businesses and organizations. On continual basis, personal or corporate websites add and/or update information on their websites to inform clientele on their up-to-date services. In short, data generation is on the ascendency and prudent methods ought to be adopted to make do of the looming data explosion situation.
Scientists including statisticians and mathematics have modeled complex approaches for interpreting trends and hidden heuristics in seemingly large data repositories. In addition to the numerous statistical and mathematical methods, symbolic approaches have been included to augment intelligent data analysis. In recent past, biologically inspired approaches have been adopted to inculcate human-like tendencies into machines for better performance. The basic reason being the imitation of the highest form of information processing – i.e. natural information processing. Since the implementation of biological approaches, they have proven better candidates in both simple and complex problem learning. Their ability to adapt to changing environmental conditions and use human-like tendencies to affect knowledge acquisition, make them unique in performance.
About the Book Chapter
In this book chapter, a closer look at biologically inspired models (Artificial Neural Networks) is done. In particular, supervised learning neural based methods are discussed within the context of directed graphs. The structural design, training, validation and testing regimes of selected feed-forward based neural networks are explained. In basic English language terms, a step by step explanation of each network is given. The underpinning mathematical rigor of each network is lessened for easy reading and comprehension. Prior to introducing each feed forward network family, motivation is provided along research lines to deepen the understanding of concepts. In terms of comprehension, the book chapter is that of undergraduate computer science study. Post-graduate students/ tutors/industry professionals/lecturers and/or researchers who wish to study or research into the foundational representations, concepts and theory of these bio-inspired models will also find this book chapter beneficial.
Arrangement of Book Chapter
The book chapter is arranged as follows. A concise literature review on search methods narrowing down to biologically plausible methods (e.g. Artificial Neural Networks (ANN)) is given. A linkage between ANN is made to signal flow graphs, which in turn, has indirect connection to graph theory. The connection is so established so as to justify that, graph theory is a fundamental theory in neuronal computation and learning. A brief introduction to directed graphs, especially signal flow graphs is given. Lastly, research motivation types (if any), training regime, validation regime, testing regime and measurement of different artificial neural network family, is discussed. At each stage, the Graphical aspects of each model are also mentioned.