神经网络课件神经网络课52教案.ppt

文档介绍：…12m…a1W2w11w21W1w11b112b2bmma2anw11LayerALayerB12n…12m…x1x2xny1y2ymLayerAW1LayerBW2x(t)=fy(fx(x(t-1)W1)W2)y(t)=fx(fy(y(t-1)W2)W1)w1fx(x(t-1)w1)fy(y(t-1)w2)w2x(t-1)x(t)y(t-1)Calculatingprocessw1fx(x(t-1)w1)fy(y(t-1)w2)w2y(t-1)x(t-1)y(t)Assuming:x{-1,+1}N,y{-1,+1}M(xi,yi),i=1,2,…,PLearningruleDefinitionofenergyfunction:wiistheithrowweightfactorsintheWwjisthejthcolumnweightfactorsintheWIfxi>0thenywiT>0Ifxi<0thenywiT<0Therefore,E<0HebblearningformularW1=W2TExample1x1(10101),y1(1111)x2(10100),y2(0110)x3(01011),y3(1001)Changethevectorsto{-1,+1}x1’(1-11-11),y1’(1111)x2’(1-11-1-1),y2’(-111-1)x3’(-11-111),y3’(1-1-11)sgn(x1W)=sgn(1551)=y1sgn(y1WT)=sgn(4–44–44)=(1–11–11)changetobinarysgn(y1WT)=(10101)=x1=(01000)x1+=(11101)sgn[(x1+)W]=sgn[2222]=-worksimplementsfuzz--logicSystemsFuzzylogicgrewoutofadesiretoquantifyrule-,often,rarely,several,few,-work————fuzzy-umericalInformationframeworkknowledgetype(basedonrules)(notamember)and1(definitelyamember).-LogicVariablesHowtoconvertanumericvariabletoafuzzy-logicvariablethroughfuzzifier,andtoconvertthefuzzy-(x)andthefuzzy-logicvariablesaredenotedfuzzy(.,50to70forlow,and80to100formoderate).(60)=fuzzy(0,1,0,0,0)numeric(90)=fuzzy(0,0,1,0,0)Fornumericvaluesinthetransitionregions,,wecanconvertthefuzzy-lo