Computational intelligence (CI) is a set of nature-inspired methodologies that, according to the editor, include “fuzzy logic (as a more human-oriented approach to reasoning); artificial neural networks (mimicking the human brain); evolutionary algorithms (mimicking the population-based genetic evolution); and dynamically evolving systems based on the above” [p. xi]. CI and artificial intelligence (AI) both deal with intelligent systems, but their approaches differ. While CI takes a more heuristics-driven approach to pattern matching via classifiers, AI is more oriented around predicate logic with an emphasis on static knowledge representation.

The two-volume handbook is divided into five parts: “Fuzzy Logic,” “Artificial Neural Networks [ANN] and Learning Systems,” “Evolutionary Computing,” “Hybrid Systems,” and “Applications.”

Part 1 includes seven chapters on fuzzy logic, which provide an in-depth look at CI’s core areas. Chapter 1 provides the framework of fuzzy sets; this excellent exposé is sufficient for a neophyte to understand how fuzzy systems work. Chapter 2 on granular computing looks at aggregation as a starting point to combine algorithmic and nonalgorithmic information processing. Chapter 3 addresses evolving fuzzy systems that capture emerging properties as part of an ongoing online mode of processing, as opposed to the batch processing mode of the 1980s. Chapters 3 and 4 lay out different approaches to fuzzy modeling. The practical information in both chapters combined enables the reader to apply fuzzy modeling to real-world systems. Chapter 5 deals with fuzzy classifiers and how uncertainty in real-world settings can be captured heuristically. Chapters 6 and 7 discuss issues of fuzzy systems as they relate to control and diagnostics, respectively.

Part 2 on ANN and learning starts in chapter 8 with a chronological overview of analyzing brain-mind processes based on mathematics. We make sense of the world by “matching concept-models to events in the world” [p. 289]. This matching process as well as the frameworks used are subject to different theories. Researchers differ in the number and nature of layers required for learning to take place. The chapter emphasizes the importance of dynamic logic (DL) to reconcile “potentialities” with “actualities” [p. 290]. Contrast DL with multiple hypotheses testing (MHT): while MHT starts out with a fixed assignment of data to a model and then tries to find results without getting lost in the combinatorics, DL starts out without any a priori commitment to the outcome but arrives at an outcome iteratively.

Chapter 9 provides a further analysis of cognitive systems with an emphasis on language and memory. A discussion of three predominant cognitive models includes symbolic, connectionist, and hybrid approaches and the computational frameworks built around them. While the field of cognitive modeling appears to be fragmented and incomplete, chapter 9 succeeds in showing the issues, potential solutions, and future ramifications of cognitive modeling. The authors in chapter 10 apply ANN to economics modeling. Since economists are usually in the business of forecasting, they start out with very limited static data but must deal mostly with dynamic systems. Often market forecasts include world models. Chapter 10 goes on to describe the application of ANN mathematics to business economics, macroeconomics, and finance.

Chapter 11 goes deeper into one derivative of the connectionist approach to cognitive modeling: evolving connectionist systems (ECOSs). ECOSs combine ANN with fuzzy rules to enable adaptive learning that is applied to medicine, health, informatics, and robotics. Chapter 12 analyzes how adaptive controllers can be optimized using reinforcement learning for “optimal control solutions online in real time by measuring data along the systems trajectories” [p. 435]. Chapter 13 is dedicated to kernel models and its most prevalent implementation, the support vector machine (SVM). Kernel methods are algorithms for pattern analysis (for example, cluster, classifications, and so on) on datasets. They only require a user-specified kernel (for example, a similarity function of raw data point representation) as opposed to non-kernel algorithms that must be normalized into feature vector representations. Kernels can add non-linearity to linear models without impact on the existing infrastructure. This process, referred to as “kernel trick,” is computationally less costly and labor intensive since it does not require computing the coordinates of a dataset but “allows replacing inner products in the linear model with kernel function values” [p. 501]. This chapter demonstrates the importance and usefulness of kernels and SVM in CI.

Part 3 covers evolutionary computing (EC) in four chapters. Chapter 14 explains the foundation, history, and philosophy of EC. Starting with a review of evolutionary theories and genetics, the three main paradigms in EC (that is, evolution strategies, evolutionary programming, and genetic algorithms) are discussed. Chapter 15 surveys artificial immune systems derived from the study of biological immune systems. Chapter 16 is a survey of swarm intelligence, including particle swarm optimization, colony optimization, bee swarm optimization, and the bat algorithm. Memetic algorithms are covered in chapter 17. The authors draw a parallel between genes used to construct a body and memes to construct a culture. While genes and memes differ in how they manifest themselves--that is, genes in our DNA, memes in language--they can be represented as strings of their respective alphabets. Hence, their evolution can be studied in EC the same way.

Part 5 discusses in four chapters real-world applications of CI: modeling human decision making (chapter 23), applying CI to the process industry in general and chemical engineering processes in particular (chapter 24), using CI for localization and pathing of robots and autonomous systems (chapter 25), and making CI part of self-driving vehicles ranging from diagnostics to navigation (chapter 26).

For anyone interested in CI with a working knowledge of set theory and applied mathematics, I would recommend this handbook. Three aspects of the handbook stand out (besides the high quality of each chapter):

(1) It avoids drawn-out philosophical discussions about any deep biological justification of intelligence. The question, “Do computers think like humans?” is vacuous and irrelevant if one wants to only appreciate what algorithms can accomplish on their own merits.

(2) The depiction of the various technologies demonstrates their complementary evolution. The handbook illustrates how the theories turn into methodologies, and then turn into actual applications.

(3) This is a “just the facts” handbook. Each author sticks to presenting the history and state of the art, along with applications of the technologies without editorializing. Ordinarily, the purpose of a handbook is to provide readers with the ability to study self-contained individual chapters related to their interests. The editor succeeded in going beyond that mode of reading and turning this into a book one wants to read from cover to cover.