Computational intelligence (CI) was in real need of this compendium, as it is an exhaustive summary that explores the theory of intelligence through possible computational means. CI, which is often confused with artificial intelligence (AI), is the modern technical description of AI. Traditional AI is but a subfield of CI, just as it is presented in this book. CI is concerned with mimicking all aspects of natural intelligence, including evolution, multicellular organisms, nervous system, immune system, swarms, and self-organization, and not just logic, reasoning, connectionism, and other such AI approaches. Furthermore, CI takes advantage of technological advancements and available resources to achieve its goal--solving real-world problems.
It is very important to note that the balance between theory and application is sustained throughout the volume. The book has more than 1,000 pages and is organized into 25 chapters, divided into ten parts. Almost every chapter contains hundreds of references, in addition to links for research groups, special interest groups, key books, open-source projects, international conferences, and workshops. Essential theoretical approaches and the latest advancements in the field are studied with corresponding real-world, challenging applications.
Obviously, Part 1 is an introductory course that presents the basic literature on the topic. Part 2 is concerned with data presentation and preprocessing, which is an important step in finding solutions to hard problems. Part 4 tackles the classic logic and reasoning area of AI. It presents a very detailed analysis of extended vector annotated logic program with strong negation (EVALPSN), an algorithm well suited for safety verification and control, applied in railway interlocking, airway traffic control, and other interdisciplinary applications. Natural language processing (NLP) is also treated in this part, as it is a traditional problem in AI. To target the problem, the book dedicates a full chapter to data-oriented parsing (DOP). DOP is a corpus-based parsing approach, based on probability and the frequency of appearance of previously parsed data. The main idea is to extract structure from a given corpus, disregarding any preset rules forming the corpus. DOP+ is a newer version of the algorithm that can, in addition, maintain simplicity and likelihood over the content of the corpus. The technique is explained for language parsing, music analysis, and problem solving. In addition, unsupervised learning approaches, very relevant in NLP, are suggested, to increase the efficacy and usefulness of parsing.
Using a classic presentation style, the compendium briefly discusses emotional intelligence and ontologies--in knowledge engineering--in Parts 3 and 5. Part 4, the most valuable part of the book, presents the core of CI, linking it with intelligent agents, fuzzy systems, artificial neural networks (ANNs), evolutionary computing, DNA computing, and immunity-based computing.
On the intelligent agents subject, an introduction is given for artificial life; more interesting yet is their application in grid computing, sensor networks, and computational economics.
As expected, the book gives utmost attention to neural networks; approximately 40 percent of the content deals with ANN research. ANNs are introduced with an interesting real-world application: media grids. How do ANNs work in media grids such as YouTube and other streaming applications? Traditional multilayer perceptrons (MLPs) are discussed and implemented as a tool for network bandwidth prediction. The MLP is trained on historical data of incoming and outgoing network traffic between media servers; the system will predict bandwidth consumption and will determine the optimal streaming strategy concerning resolution selection and job scheduling across different servers. The application is presented with system analysis, experimental results, and performance evaluation. ANNs are then further studied through the exposition of evolutionary neural networks, self-organizing maps (SOM), and neural systems engineering. Furthermore, genetic algorithms that solve set partitioning problems, such as graph coloring and timetabling problems, genetic programming, and the particle swarm algorithm are fully explained.
The last part of the volume is concerned with DNA and immunity-based computing. In DNA computing, data is represented using DNA sequences. The main idea is to use DNA molecules for information storage and to manipulate execution via special chemical reactions. In this way, an increase in speed is achieved using the default property of parallelism that occurs in a mixture of DNA molecules contained in a test tube. Storage capacity is also massively increased, since data is represented at the molecular level. Additional electronic devices are placed over the test tube for encoding and visualization. DNA computing can be used to solve nondeterministic polynomial time (NP)-complete and NP-hard problems; an example is given and studied for elevator scheduling. Immunity-based computing was inspired by immune system mechanisms such as self-defense and self-maintenance, which are implemented within the DNA computer.
The book has much more valuable content; I can’t address all of it in this review. This compendium is essential reading for researchers with an interest in the field; it could also serve as a resource for a doctoral thesis or various research projects, as it refers to literature that covers every aspect of the field and provides promising solutions to CI open problems.
This compendium would be the last word in the history of CI, except it doesn’t include the topics of chaotic neural networks and quantum computing, which show much promise in the field. It is possible that the volume will have future versions, after a unifying theory linking all these fields is found, and then the limits of AI will finally be revealed.
