A Graph-Theoretic Analysis of Information Value
Kim-Leng Poh, Eric Horvitz
We derive qualitative relationships about the informational relevance of variables in graphical decision models based on a consideration of the topology of the models. Specifically, we identify dominance relations for the expected value of information on chance variables in terms of their position and relationships in influence diagrams. The qualitative relationships can be harnessed to generate nonnumerical procedures for ordering uncertain variables in a decision model by their informational relevance.
Keywords: Expected value of information, graphical analysis, d-separation,
PS Link: ftp://ftp.research.microsoft.com/pub/ejh/gev.ps
PDF Link: /papers/96/p427-poh.pdf
AUTHOR = "Kim-Leng Poh
and Eric Horvitz",
TITLE = "A Graph-Theoretic Analysis of Information Value",
BOOKTITLE = "Proceedings of the Twelfth Conference Annual Conference on Uncertainty in Artificial Intelligence (UAI-96)",
PUBLISHER = "Morgan Kaufmann",
ADDRESS = "San Francisco, CA",
YEAR = "1996",
PAGES = "427--435"