دانلود کتاب هوش مصنوعی: اصول و تمرین

This book offers a comprehensive introduction to the exciting and too often mysterious discipline of Artificial Intelligence. As an introduction to AI, it covers the foundational computational technologies that have supported work in AI since its inception over 70 years ago. In Part I, it covers the mathematical, philosophical, and engineering disciplines that make artificial intelligence possible. AI is not a standalone discipline but rather integrates and expands the many insights of the psychologists, mathematicians, neuroscientists, and others that are part of its current research and practice. The final chapters, Part VIII, focus on the limitations, ethical issues, and future promise of AI. Here we present many of the social issues that must be considered when integrating AI technology into our modern world. There are three major design paradigms that have emerged in AI research and engineering over the past 70 years. The first paradigm, begun in the 1950s, is the symbol-based approach to problem-solving. Symbol-based AI is founded on graph theory, search algorithms, and the use of representational techniques including semantic networks and the propositional and predicate calculi. These technologies are covered in Parts II through V. The second paradigm of AI practice is neural or connectionist network problem-solving. Although this tradition had its roots in the 1940s and was mentioned in the AI manifesto made at the Dartmouth summer workshop in 1956, it did not reach its full potential until the creation of the Boltzmann machine and the backpropagation algorithms of the 1970s and 1980s (Werbos 1994; Rumelhart and McClelland 1986). Since then, neural networks and deep learning have produced many quality results in image processing, human language analysis, and other areas of complex information processing. Neural network problem-solving, with an introduction to deep learning, is covered in the six chapters of Part VI. Probabilistic or stochastic approaches to problem-solving make up the third major paradigm of AI. This approach uses Bayesian-based probability measures to address the uncertainties and impreciseness of world situations. Probabilistic AI began in the digit and image recognition studies of the 1950s, but only reached large-scale acceptance with successes in human language processing of the 1990s. Pearl’s (1988, 2000) insights, seen in the creation of Bayesian belief networks and dynamic Bayesian networks, allowed the many forms of Bayesian reasoning to be computationally tractable, more transparent, and accountable to the user. Probabilistic models are presented in Part VII. There are other minor paradigms, with much more limited use in current AI practice. These include genetic algorithms, artificial life, fuzzy and other logic-based forms for uncertain reasoning. Genetic algorithms, suggested by Holland in the 1970s, and emergent computation, first proposed by von Neu-mann in the 1940s, reflect Darwin’s insights on population growth and survival of the fittest members of a society. These techniques are presented in subsections of Part II. Fuzzy and other logic-based schemes for reasoning in uncertain situations are presented in 7 Chap. 12. The principles and practice of these paradigms make up the foundation and successes of work in current artificial intelli-gence. They also reflect the structure and the recommended uses of our course materials.