Qualitative Methods for Reasoning under Uncertainty
The problem of reasoning under uncertainty is widely recognised as being of great importance in artificial intelligence. A wide range of formal methods have been proposed for dealing with the problem, and there has been much debate in the literature as to which formalism is best. In recent years several researchers have adopted an eclectic position proposing that the different formalisms are in fact complementary, providing means of modelling different nuances of uncertainty. Such a position leads naturally to the use of several techniques in the solution of a single problem, and hence to the problem of combining the results of applying these techniques.
This book addresses the problem of integrating different formalisms. It is proposed that a corollary of the eclectic approach is that any value in any formalism has the same underlying meaning of ``a reason to believe that an event might occur''. This provides two means of integrating values between formalisms. In the first approach methods from qualitative reasoning are used to express unknown values and combine these with more conventional point values. Any value from one formalism may then be translated into another formalism as an unknown value and combined with values in the second formalism.
In the second approach, qualitative methods are used to analyse probability, possibility and evidence theories so that it is possible to predict the way in which changes in value are propagated in those formalisms. As a result it is possible to describe how changes that result from new evidence, are propagated uniformly across several formalisms. An implementation of this work based on local computation and qualitative algebras is described, and the implementation is applied to a problem from the complex domain of protein topology prediction.
The book is a substantially revised and expanded version of my thesis. I maintain a list of errata