This second edition includes new information and some minor error corrections to the previous edition [1]. The content is structured into three parts devoted to probabilistic reasoning, learning causal models, and knowledge engineering.
The five chapters of Part 1 present the fundamentals, concepts, models, and results currently used in probabilistic reasoning. The authors briefly discuss the basics of reasoning under uncertainty in the first chapter. The second chapter presents the structure of Bayesian networks, as well as Pearl’s network construction algorithm and the principles of reasoning with Bayesian networks. It also provides quick guides to using the dedicated software BayesiaLab, GeNIe, Hugin, and Netica. The next chapter provides a detailed treatment of the problem of belief updating in Bayesian networks. The main topics that this chapter presents are Kim and Pearl’s message passing algorithm, belief updating based on uncertain inference, inference in multiply-connected networks, and approximate inference based on stochastic simulation.
The fourth chapter is devoted to decision networks resulting from including utilities assigned to different possible outcomes and extending Bayesian networks to support decision making. Section 4.5 introduces the generalization of Bayesian networks, called dynamic Bayesian networks (DBNs), together with the principles of reasoning and inference algorithms for DBNs, in order to explicitly model change over time. The final section of the chapter introduces object-oriented Bayesian networks as a suitable framework for building large, complex, hierarchical Bayesian decision networks. The final chapter of this part presents several applications in medicine and ecology.
Part 2 addresses the problem of applying machine learning techniques to the task of learning Bayesian networks from data. Chapter 6 presents the problem of learning from data the parameters representing the entries of the conditional probability tables of discrete networks, together with suitable algorithms, such as the Gibbs sampling method and maximum likelihood expectation maximization. The next three chapters present the Bayesian classifiers, including naive and semi-naive Bayes models, and ensemble Bayes classification; several strategies for evaluating the resulting performance; and the problems of learning linear models and different discrete causal structures.
The final part of the book comprises two chapters. These chapters present key aspects in Bayesian network-based knowledge engineering, including case studies. The list of bibliographic references contains a significant number of new and older titles of representative published work.
This well-written, outstanding book will be extremely useful to a broad range of readers, including computer science students, postgraduate students, academic researchers, and people involved in the development of applications of knowledge processing in various domains. The authors manage to assure both a rigorous treatment from a mathematical point of view, and accessibility. The well-selected illustrative examples lead the reader toward a deeper understanding of the concepts, models, and algorithms.