Construction of preferred causal hypotheses for reasoning with uncertain knowledge
Raj Bhatnagar, Laveen N. Kanal · 1989
This dissertation presents a framework for reasoning by constructing causal hypotheses to explain the observed events. We develop a hypergraph based formalism for representing the knowledge of known causal relationships of a domain. With each known causal relationship, we associate a probability distribution to represent the uncertainty associated with this knowledge. We then define a scenario as a hypothesis which includes a subset of the known causal relationships such that this subset explains the observed events and is consistent according to certain specified criteria. Since it is possible to construct a number of hypotheses to explain the same set of observations, we use a criterion of interestingness to select one out of the possible hypotheses. This criterion depends on the inferences that can be made about some aspect of interest in the context of the hypothesized scenario. Each scenario is a subgraph of the complete domain knowledge hypergraph. To construct a scenario that satisfies the interestingness criterion we need to identify the appropriate subgraph of the complete domain knowledge hypergraph. We present an algorithm based on an artificial intelligence search procedure to identify the interesting scenarios for various types of interestingness criteria.