Uncertainty in Artificial Intelligence

spirit, Fu (p. 119) presented a method of nonmonotonic reasoning in a neural net representation. A theme from the 1987 workshop, control of inference processes, is developed both as a method of limiting combinatorics and as an inherent component of AI systems. D’Ambrosio (p. 64) and Jimison (p. 189) presented methods for modular, interactive system modeling, inference, and control. Utility theory provides a control approach for systems whose uncertain inference is represented in a Bayesian framework. Levitt, Binford, et al. (p. 245), Breese and Fehling (p. 30), Hansson and Mayer (p. 148), and Horvitz (unpublished) provided implementation techniques and efficiency arguments for utility-based control. Pittarelli (p. 283) and Wen (p. 360) presented similar arguments about minimum entropy approaches. Paul Cohen (unpublished) gave a persuasive presentation for the integration of procedural (for example, reactive) representations, with numeric calculi, in the context of an automated planner for fire fighting control. Jain and Agogino (p. 178) and Eick (p. 98) presented fuzzy approaches to the representation and control of systems. Here, the emphasis was as much on issues of the sensitivity of variables in presentations and their influence on control of systems as it is was on the dichotomy between fuzzy and probabilistic representations. The comparison and fusion of multiple UAI theories continued to be central issues. Neapolitan and Kenevan (p. 266) compared classical to Bayesian probability; Kwok and Carter (p. 213) compared averaging effects in decision trees, and Kalagnanam and Henrion (p. 205) compared Bayesian to heuristic techhe Fourth Workshop on Uncertainty in Artificial Intelligence was held 19–21 August 1988 at the University of Minnesota. The workshop featured significant developments in the application of theories of representation and reasoning under uncertainty to diverse areas. These areas included automated planning, temporal reasoning (Dean and Kanazawa, p. 73; Dutta, p. 90*), computer vision (Agosta, p. 1; Levitt, Binford, et. al., p. 245), medical diagnosis (Cecile et al., p. 38), fault detection (Gallant, p. 127), text analysis (Tessem and Ersland, p. 344), distributed systems (Frisch and Haddawy, p. 109), and behavior of nonlinear systems (Yeh, p. 374). A central thrust that uncertainty in artificial intelligence (UAI) brings to these applications is the pressing need to handle real-world combinatorics. Here, UAI techniques can make a unique contribution to the AI world. For example, Kyburg (p. 229) suggested that nonmonotonic reasoning might be subsumed by appropriately formulated probabilistic inference, bypassing the combinatorics inherent in purely logical representations; Yager (p. 368) presented a similar scheme for possibility theory. Neufeld and Poole (p. 275) suggested that an ordinal, qualitative calculus, might be sufficient to deal with nonmonotonic and default representations. However, the theoretical results of Aleliunas (p. 8) were evidence that calculi with any ordinal value space might be forced to pure probability. Kadie (p. 197), Bacchus (p. 15), and Frisch and Haddawy (p. 109) also presented methods for integrating probabilistic and logical reasoning, both to limit combinatorics and maximize the expressiveness of representations. In the same The Fourth Uncertainty in Artificial Intelligence workshop was held 19–21 August 1988. The workshop featured significant developments in application of theories of representation and reasoning under uncertainty. A recurring idea at the workshop was the need to examine uncertainty calculi in the context of choosing representation, inference, and control methodologies. The effectiveness of these choices in AI systems tends to be best considered in terms of specific problem areas. These areas include automated planning, temporal reasoning, computer vision, medical diagnosis, fault detection, text analysis, distributed systems, and behavior of nonlinear systems. Influence diagrams are emerging as a unifying representation, enabling tool development. Interest and results in uncertainty in AI are growing beyond the capacity of a workshop format. T