This paper is concerned with a development of a theory on probabilistic models, and in particular Bayesian networks, when handling continuous variables. While it is possible to deal with continuous variables without discretisation, the simplest approach is to discretise them. A fuzzy partition of continuous domains will be used, which requires an inference procedure able to deal with soft evidence. Soft evidence is a type of uncertain evidence, and it is also a result of the type of discretisation used. An algorithm for inference in multiply connected networks will be proposed and exploited for filtering and abduction in dynamic, time-invariant models, when continuous variables are present.
An approach to hybrid probabilistic models
2008
Abstract
This paper is concerned with a development of a theory on probabilistic models, and in particular Bayesian networks, when handling continuous variables. While it is possible to deal with continuous variables without discretisation, the simplest approach is to discretise them. A fuzzy partition of continuous domains will be used, which requires an inference procedure able to deal with soft evidence. Soft evidence is a type of uncertain evidence, and it is also a result of the type of discretisation used. An algorithm for inference in multiply connected networks will be proposed and exploited for filtering and abduction in dynamic, time-invariant models, when continuous variables are present.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
