TowardsRigorousInterpretations:aFormalisationofFeatureAttributionDariusAfchar12RomainHennequin1VincentGuigue2Abstractcontrollinganddisentanglingeveryaspectofamodelpre-diction,rangingfromhowthedatai...
Flow-basedAttributioninGraphicalModels:ARecursiveShapleyApproachRaghavSingal1GeorgeMichailidis12HoiyiNg1AbstractForinstance,asweillustratein§6,ourframeworkcanbeusedtounderstandwhethersensitiveattr...
BiasAlsoMatters:BiasAttributionforDeepNeuralNetworkExplanationShengjieWang1TianyiZhou1JefferyA.Bilmes2Abstractfunctions,utilizingmulti-layerarchitecturesandarichsetofpossiblehiddennon-linearities,m...
InterpretabilityBeyondFeatureAttribution:QuantitativeTestingwithConceptActivationVectors(TCAV)BeenKimMartinWattenbergJustinGilmerCarrieCaiJamesWexlerFernandaViegasRorySayresAbstractcoefficientweigh...
AxiomaticAttributionforDeepNetworksMukundSundararajan1AnkurTaly1QiqiYan1AbstractShrikumaretal.,2016;Binderetal.,2016;Springenbergetal.,2014).Westudytheproblemofattributingthepre-dictionofadeepnetwo...