Graphical Analysis of Value of Information in Decision Models

Wereviewand extendthe qualitative relationships about the informational relevanceof variables in graphical decision modelsbased on conditional independenciesrevealedthroughgraphicalseparations of nodesfromnodesrepresentingutility onoutcomes. Weexploitthesequalitativerelationshipsto generate non-numericalgraphical proceduresfor identifying partial orderingsoverchancevariablesin a decision modelin termsof their informationalrelevance.We describean efficientalgorithmbasedona consideration of local properties of a propertywerefer to as useparation.Finally,wepresentresultsof computational efficiencies gainedvia the application of the new policies, basedon analysesof samplenetworkswith differentdegreesof connectivity.

• Type: Conference
• Year: 2001
• Where: FLAIRS
• Authors: Songsong Xu, Kim-Leng Poh, Eric Horvitz
• Keywords: Artificial Intelligence, FLAIRS 2001, Graphical Decision, Informational Relevanceof Variables, Useparation.Finally