In the 1980s, researchers from
cognitive science (e.g.,
Judea Pearl),
computer science (e.g.,
Peter C. Cheeseman and
Lotfi Zadeh), decision analysis (e.g.,
Ross Shachter), medicine (e.g.,
David Heckerman and
Gregory Cooper), mathematics and statistics (e.g., Neapolitan,
Tod Levitt, and
David Spiegelhalter) and philosophy (e.g.,
Henry Kyburg) met at the newly formed Workshop on Uncertainty in Artificial Intelligence to discuss how to best perform uncertain inference in artificial intelligence. Neapolitan presented an exposition on the use of the classical approach to probability versus the Bayesian approach in artificial intelligence at the 1988 Workshop. A more extensive philosophical treatise on the difference between the two approaches and the application of probability to artificial intelligence appeared in his 1989 text
Probabilistic Reasoning in Expert Systems: Theory and Algorithms. The text defines a causal (Bayesian) network, and proves a theorem showing that a
directed acyclic graph G and a discrete probability distribution P together constitute a Bayesian network if and only if P is equal to the product of its conditional distributions in G. The text also includes methods for doing inference in Bayesian networks, and a discussion of influence diagrams, which are Bayesian networks augmented with decision nodes and a value node. Many AI applications have since been developed using Bayesian networks and influence diagrams. Neapolitan's "Probabilistic Reasoning in Expert Systems" have been widely recognized as formalizing the field of Bayesian networks, as seen in the works of
Eugene Charniak, who, in 1991, noted both texts as the source for Bayesian network inference algorithms; P.W. Jones, who wrote a review of "Probabilistic Reasoning in Expert Systems"in 1992; Cooper and Herskovits, who credit Neapolitan's text and Pearl's text for formalizing the theory of belief networks in their 1992 paper that developed the score-based method for learning Bayesian networks from data; and Simon Parsons, who, in 1995, compared the two texts and discussed their roles in establishing the field of probabilistic networks. More recently, in 2008, Dawn Holmes discussed Neapolitan's career and the contribution of his first text. which applies Bayesian networks to problems in finance and marketing; and
Probabilistic Methods for Bioinformatics, which applies Bayesian networks to problems in biology. Neapolitan has also written
Foundations of Algorithms and (with Xia Jiang)
Artificial Intelligence: With an Introduction to Machine Learning. == References ==