Knowledge Updating in Bayesian Networks

The problem of reasoning in Bayesian networks with knowledge updating is considered. The reasons of this updating are the following: not precise data evaluation, changes in the surrounding world, testing data introduction and adaptive systems. This knowledge updating forces the repetition of the reasoning process. The idea of updating process in message-passing architectures is presented. The computational complexity of passing of updating messages is explored. Some remarks on the improving of the updating process and the optimisation problem are presented. The motivation of designing specialised heuristics is pointed out. Local search heuristic based on the idea of evolutionary programs is proposed.