A probabilistic and informational basis to optimize expert systems

This paper presents a knowledge representation and a probability estimation basis for a diagnostic oriented expert system. The paper also includes a dynamic programming approach and an informational heuristic to optimize the related query process. The authors emphasize the interface between the phase of knowledge acquisition and representation of the probabilistic and logical relations. In this sense, they show how to cope with information under uncertainty, in order to estimate an adequate value for the probability of the occurrence of any possible diagnostic. This results in a optimization problem, where the constrains contain the provided information, and where an entropy maximization criterion search for non-tendentious estimates. The authors discuss the use of appropriate techniques to solve such types of mathematical programming problems, with emphasis on the possibility of parallel computation, by using duality decomposition in large-scale versions with strictly concave maximization criteria and linear equations. The importance of the mathematical programming model is related with the central role of the output estimates to optimize the query process of a diagnostic oriented expert system, which is also discussed in the paper.