Seminar: Wen Yan, "The CPT Structure of Variable Elimination in Discrete Bayesian Networks", November 15th , 12:30 pm, CL 417 (Expired)

Department of Computer Science




SPEAKER:     Wen Yan


DATE:             November 15, 2010


TIME:             12:30 pm        


PLACE:          CL 417


TITLE:            The CPT Structure of Variable Elimination in Discrete Bayesian Networks





The task of variable elimination (VE) is central to reasoning with Bayesian networks. A variable v is eliminated using a simple two-step procedure. First, the probability distributions involving v are multiplied together. Second, v is marginalized out of the product obtained in the first step. Other variables can be eliminated in a recursive manner. The probabilistic reasoning literature has always denoted the probability distributions constructed during VE as potentials. This description is not as precise as it should be.


In this talk, I will show that every multiplication operation and every marginalization operation involved in eliminating variables from a discrete Bayesian network yields a CPT. The significance of this result resides in the description of the VE algorithm. Potentials do not have clear physical interpretation, as they are unnormalized probability distributions. In contrast, CPTs have clear semantic meaning, since the probabilities in the distribution must necessarily obey a specific pattern. Thereby, establishing that the distributions constructed by VE are, in fact, CPTs rather than potentials yields a description that is more precise and readable.







Breaking News

Recent News

Do You Have News for the Department of Computer Science?

  • Please send a plain text version of your posting, which can include URL links, to
To Top of Page