Department of Computing, University of Bradford
University Information Service: Contents | Index | Search

Reasoning under Uncertainty


Research by Dr P. Rhodes and Mr G. Garside

Dr Rhodes, Mr Garside and two research students are doing research into the use of probability for Reasoning under Uncertainty. Their aim is to extend the applicability of causal networks to situations in which the available information is incomplete or not in a suitable form. This is being achieved by using whatever causal information is available to find the global maximum entropy solution and hence a minimally prejudiced estimate of the information required by the causal network. The global maximum entropy solution, hitherto thought to be exponentially complex, has been shown to be obtainable by local computations in linear time for certain classes of causal networks. For some of these classes, the traditional Lagrangian method for finding the global maximum entropy solution has had to be extended to handle non linear constraints arising from independence between events. Algorithms for computing minimally prejudiced causal information have been developed using these results. Future research will attempt to extend these results to all classes of causal network. When this has been done, it will be possible to produce a pre-processor for converting incomplete information into a form which can be used by current packages based on causal networks, eg HUGIN. A longer term aim is to extend the work beyond the formal structures of causal networks with the intention of applying it to decision making which, in most practical circumstances, has to be done with incomplete causal and non causal information.

Publications


Author: P.Rhodes
Last update: 22/1/96

Up to AI group homepage