Independence in Uncertainty Theories and Its Applications to Learning Belief Networks

Data Mining with Graphical Models

“…In this case the conditional independence tests, obviously, can be restricted to orders 0 and 1, because for any pair of attributes there is only one path connecting them, which can be blocked with at most one attribute. An overview of other specialized version that consider somewhat less restricted classes of graphs, but which are all, in one way or the other, based on the same principle, is given in [de Campos et al 2000]. Note that the assumption of a sparse graph is admissible, because it can only lead to additional edges and thus the result of the algorithm must be at least an independence map (although it may be more complex than the type of graph assumed, e.g., more complex than a polytree).…”

Learning Undirected Possibilistic Networks with Conditional Independence Tests

“…Several search methods and evaluation measures, which are the core ingredients of any algorithm for learning graphical models, were developed and applied not only to learning probabilistic graphical models, but also to the somewhat less well known possibilistic networks [13], [1], [8], [3].…”