In this paper, we present an approach to extract most relevant information from a (semi-)quantitative knowledge base, e.g., from a probability distribution. Relevance here is meant with respect to some appropriate inductive inference process, like maximum entropy inference (ME-inference) in probabilistics. So in particular, the method developed in this paper is apt to solve the inverse maxent problem, computing from a distribution in a non-heuristic way a set of conditionals that ME-represents that distribution. Since we only make use of one special characteristic of ME-inference, this method may as well be applied to other, similar inference processes.
Keywords: Data Mining and Knowledge Discovery, Knowledge Representation, Uncertainty in AI, Conditionals, Maximum Entropy
Citation: Gabriele Kern-Isberner: Solving the Inverse Representation Problem. In W.Horn (ed.): ECAI2000, Proceedings of the 14th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2000, pp.581-585.