Apr. 4 2011 2:00PM

Apr. 4 2011 3:30PM

The theory of compositional models covers one approach for knowledge representation; it is a class of models built within the framework of probability theory. The basic idea of the approach is the same as that on which expert systems are based: it is beyond human capabilities to represent or express global knowledge of an application area - one has always to work only with pieces of local knowledge. Such a local knowledge can be, within probability theory, easily represented by an low-dimensional distribution. When pieces of local knowledge are represented by distributions, the global knowledge should be represented by a multidimensional probability distribution - a compositional model.

In this talk a problem of conditioning will be presented. This problem is closely related with the so called "flexible sequences" that are artificially created to enable conditioning in the framework of compositional models. However, our knowledge about flexible sequences (and their properties) is very poor. Several open questions, examples, and hypothesis will be discussed.

Pravidelný seminář Inteligentní systémy (Contact person: Tomáš Kroupa)