Jun. 13 2011 2:00PM

Jun. 13 2011 3:30PM

In this talk we discuss representation by means of an acyclic directed graph (DAG) of the independence model induced by a coherent T-conditional possibility (where T=min or a strict t-norm). Such a model is in general not closed under symmetric property, so we must rely on a proper asymmetric notion of vertex separation. Structures closed w.r.t. all graphoid properties (and their symmetrized versions) except for symmetry, have been characterized with the name of a-graphoids. We present an efficient method to generate and store the independence model, based on the concept of fast closure w.r.t. a-graphoid properties. Moreover, we introduce asymmetric Markov properties and prove their equivalence for a-graphoids. In this way, an algorithm to build an I-map, given an order of the random variables, can be derived. A systematic procedure to extract all independence statements encoded in a graph is also provided. Finally, we define coherent T-possibilistic networks and show how to make inference on them.

Pravidelný seminář Inteligentní systémy (Kontaktní osoba: Tomáš Kroupa)