|
|
自动化学报 2005
双向联想记忆神经网络抗噪声的能力, PP. 668-674 Keywords: Associativememory,neuralnetwork,noisesensitivity,digitalpattern Abstract: ？Thispaperanalyzesnoisesensitivityofbidirectionalassociationmemory(BAM)andshowsthattheanti-noisecapabilityofBAMrelatesnotonlytotheminimumabsolutevalueofnetinputs(MAV),assomeresearchersfound,butalsotothevarianceofweightsassociatedwithsynapseconnections.Infact,itisdeterminedbythequotientofthesetwofactors.Onthisbase,anovellearningalgorithm—smallvarianceleaningforBAM(SVBAM)isproposed,whichistodecreasethevarianceoftheweightsofsynapsematrix.Simulationexperimentsshowthatthealgorithmcandecreasethevarianceofweightsefficiently,therefore,noiseimmunityofBAMisimproved.Atthesametime,perfectrecallofalltrainingpatternpairsstillcanbeguaranteedbythealgorithm.
|
