文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Belief revision in probability theory
侵权投诉

Belief revision in probability theory

In a probability-based reasoning system, Bayes' theorem and its variations are often used to revise the system's beliefs. However, if the explicit conditions and the implicit conditions of probability assignments are properly distinguished, it follows that

PeiWang

CenterforResearchonConceptsandCognition

IndianaUniversityBloominton,IN47408pwang@cogsci.indiana.edu

Abstract

Inaprobability-basedreasoningsystem,Bayes’theoremanditsvariationsareoftenusedtorevisethesystem’sbeliefs.However,iftheexplicitcon-ditionsandtheimplicitconditionsofprobabilityassignmentsareproperlydistinguished,itfollowsthatBayes’theoremisnotagenerallyapplicablerevisionrule.Uponproperlydistinguishingbe-liefrevisionfrombeliefupdating,weseethatJeffrey’sruleanditsvariationsarenotrevisionrules,either.Withoutthesedistinctions,thelim-itationoftheBayesianapproachisoftenignoredorunderestimated.Revision,initsgeneralform,cannotbedoneintheBayesianapproach,becauseaprobabilitydistributionfunctionalonedoesnotcontaintheinformationneededbytheoperation.

Inrevision,thesystemmodi esthecertaintyvalueofanexistingproposition.

Otherauthorsmayusethetwowordsdifferently(Pearl1988),butwecanstillmaketheabovedistinction,nomatterhowtheoperationsarenamed.

Bayes’theoremseemstobeanexception.Insomesystems,itisusedasapropagationrule,whileinothers,asarevisionrule.Let’sbeginouranalysiswiththesetwousages,whichwillleadustothekernelofthedebateonthelimitationsoftheBayesianapproachasamodelofreasoningwithuncertainty.

2PROPAGATIONVS.REVISION

1INTRODUCTION

Inareasoningsystemthatdealswithuncertainty,apropo-,whereisasentencesitioncanberepresentedas

ofaformallanguage,andindicatesthesentence’suncer-tainty.

Accordingtoprobabilitytheory,itisnecessarytode neapropositionspaceinorderforasystemtorepresentandcalculateprobabilitiesofpropositions.isthesetofallpropositionstobeprocessedbythesystem,andmaybegeneratedfromasetofatomicpropositions,usinglogicaloperators(WiseandHenrion1986).

Asthestartingpointofallprobabilitycalculations,apriorprobabilitydistributionshouldbede nedon,undertheconstraintsoftheaxiomsofprobabilitytheory.

Indifferentsystems,andmayhavedifferentformsandinterpretations,andtheoperationsonthemmaybede neddifferently.However,therearestilloperationssharedbymanysystems,inspiteofallthedifferences(BhatnagarandKanal1986):

Comparison:Todecidewhichofthe

(estcertaintybycomparing

hasthehigh-12).

Tochoosesuchapriorprobabilitydistributionforaspe-ci cproblemdomain,somebackgroundknowledge(such

asstatisticaldataorsubjectiveestimates)andgeneralprin-ciples(suchastheprincipleofindifferenceormaximumentropy)arenecessary.Let’srefertothemcollectlyas,theimplicitconditionofthedistributionfunction,andwritethepriorprobabilitydistributionfunctionas

Propagation:Togetaconclusion

,whereofpremises

12.

1

1

fromaset

isdifferentfrom

1

:

01

Revision:Tomodifytheuncertaintyvalueofaproposition

fromtointhelightofotherpropositions.De nedasabove,propagation(orinference)andrevision

(orcombination)areclearlydifferent.

Inpropagation,thesystemgeneratesanewproposition(withitscertaintyvalue)thatwasnotamongthepremises.

FromBayes’theorem,wecangettheconditionalprobabil-ityof1undertheconditionthat2istrue(obviously,theconditionalprobabilityisalsobasedon):

1

2

1

2

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email