Prior knowledge for learning networks in non-probabilistic settings

Current learning methods for general causal networks are basically data-dri ven. Exploration of the search space is made by resorting to some quality m easure of prospective solutions. This measure is usually based on statistic al assumptions. We discuss the interest of adopting a different point of vi ew closer to machine learning techniques. Our main point is the convenience of using prior knowledge when it is available. We identify several sources of prior knowledge and define their role in the learning process. Their re lation to measures of quality used in the learning of possibilistic network s are explained and some preliminary steps for adapting previous algorithms under these new assumptions are presented. (C) 2000 Elsevier Science B.V. All rights reserved.