Working of Neural Networks
-
Upload
navinsehgal -
Category
Documents
-
view
222 -
download
0
Transcript of Working of Neural Networks
-
8/9/2019 Working of Neural Networks
1/12
Working of NeuralNetworks
-
8/9/2019 Working of Neural Networks
2/12
Working of a simple neuron
We selected four inputs of three different flowers(Setosa,Verginica,Versicolor)
X1-> Petal Width
X2-> Petal length X3-> Sepal width X4-> Sepal length Each neuron computes output as- Uk = 4*X1+2*X2+X3+X4
Vk=Uk + Bias Now value of Vk decides the flower.
-
8/9/2019 Working of Neural Networks
3/12
True-391084430132Setosa
True-231245535132Setosa
True-131345137154Setosa
True-61415738173Setosa
True-371104830141Setosa
True-381094429142Setosa
True-141335542142Setosa
T
rue-131345434154Setosa
True-251225036142Setosa
True-321154930142Setosa
True-41435138194Setosa
True-271205030162Setosa
True-181295138162Setosa
True-371104432132Setosa
True-301174936141Setosa
True-281194831162Setosa
True-371104636102Setosa
True-281195033142Setosa
Function(F=1 if V1
-
8/9/2019 Working of Neural Networks
4/12
True5031068325923Verginica
False-725363285115Verginica
True3229264285622Verginica
True3229268305521Verginica
True7033072366125Verginica
True3229269315421Verginica
True2028067255818Verginica
True226256284920Verginica
True4730769325723Verginica
False-1824260225015Verginica
True1727764315518Verginica
True326358275119Verginica
True2828864285621Verginica
False026063274918Verginica
True426463255019Verginica
False-2823249254517Verginica
True326358275119Verginica
True1927965305220Verginica
True3429469315123Verginica
Function(F)V1=U1-260(Bias B1)U1=4*X1+2*X2+X3+X4Sepal_length(X4)Sepal_width(X3)Petal_length(X2)Petal_width(X1)Species_Name
Neuron 2(Training Set)
Neuron 2 computations U2, Bias , V2 for flower verginica with output v1>0 andV1< 100
-
8/9/2019 Working of Neural Networks
5/12
Neuron 3 computations U3, Bias , V3 for flower versicolor with output v1>200
True-5120956293613Versicolor
True-8018051253011Versicolor
True-1424663254915Versicolor
True-3422663234413Versicolor
True-2723359304215Versicolor
True-4921158273912Versicolor
True-4721352273914Versicolor
True-8117949243310Versicolor
True-2123960294515Versicolor
True-5320758274110Versicolor
True-8117950233310Versicolor
True-4821258264012Versicolor
True-825270324714Versicolor
True-625463334716Versicolor
FunctionV1=U1-230(Bias B1)U1=4*X1+2*X2+X3+X4Sepal_length(X4)Sepal_width(X3)Petal_length(X2)Petal_width(X1)Species_Name
Neuron 2(Training Set)
-
8/9/2019 Working of Neural Networks
6/12
verginica28864285621Verginica
setosa174936141Setosa
setosa194831162Setosa
versicolor17950233310Versicolor
setosa1104636102Setosa
verginica26358275119Verginica
versicolor25270324714Versicolor
setosa195033142Setosa
Neuron 3(Test)Neuron 2(Test)Neuron1 (Test)V1(Output)Sepal_length X4Sepal_width X3Petal_length X2Petal_widthX1
Species
Classification based on training set
Applying Neuron 1 test, Neuron 2 test , Neuron 3 test to classify flowersDepending on the final output and after subtracting bias
-
8/9/2019 Working of Neural Networks
7/12
6.0170092245138194Setosa
4.5215484075030162Setosa
4.0944352485138162Setosa
6.7263957664432132Setosa
51X4(test)=2.5854206624936141Setosa
34X3(test)=4.1574511424831162Setosa
15X2(test)=6.4284057124636102Setosa
2X1(test)=1.498132175033142Setosa
1.115526781Min=D(Xtest,Xi)Sepal_length(X4)Sepal_width(X3)Petal_length(X2)Petal_width(X1)Species_Name
50.1X4(test1)=34.28X3(test1)=14.62X2(test1)=2.46X1(test1)=
Neuron 1 (Training Set)
Memory based learning to select vector X(test) for Neuron 1 by finding minimumdeviation from the mean. (Min value not in picture)
-
8/9/2019 Working of Neural Networks
8/12
9.50807981958275119Verginica
2.76927571164285621Verginica
7.86334237363274918Verginica
7.8555524563255019Verginica
65X4(test)=20.6332283649254517Verginica
30X3(test)=9.50807981958275119Verginica
55X2(test)=3.62467594765305220Verginica
18X1(test)=6.40276647869315123Verginica
2.235043305Min=DeviationSepal_length(X4)Sepal_width(X3)Petal_length(X2)Petal_width(X1)Species_Name
65.857143X4(test2)=29.71428571X3(test2)=55.51020408X2(test2)=19.97959184X1(test2)=
Neuron 2(Training Set)
Memory based learning to select vector X(test) for Neuron 2 by finding minimumdeviaton from the mean.
-
8/9/2019 Working of Neural Networks
9/12
8.56294526752273914Versicolor
15.3588886449243310Versicolor
2.92029335660294515Versicolor
4.22064302858274110Versicolor
60X4(test)=14.9781320550233310Versicolor
29X3(test)=4.04787056958264012Versicolor
45X2(test)=12.0981340770324714Versicolor
15X1(test)=7.98452272163334716Versicolor
2.920293356Min=D(Xtest,Xi)Sepal_length(X4)Sepal_width(X3)Petal_length(X2)Petal_width(X1)Species_Name
59.408163X4(test3)=27.63265306X3(test3)=43.18367347X2(test3)=13.26530612X1(test3)=
Neuron 3(Training Set)
Memory based learning to select vector X(test) for Neuron 3 by finding minimumdeviation from the mean.
-
8/9/2019 Working of Neural Networks
10/12
11.53846154Error %=
verginica13.190905963.87298334647.4025315864285621Verginica
setosa36.428011247.56048783.162277664936141Setosa
setosa34.0293990545.464271694.3588989444831162Setosa
versicolor17.464249228.6705423722.5831795850233310Versicolor
setosa40.4845649651.749396137.3484692284636102Setosa
verginica7.7459666928.66025403841.0243829958275119Verginica
Versicolor10.6770782510.4403065139.153543970324714Versicolor
Setosa35.2987251946.593991031.7320508085033142Setosa
MIN DeviationDeviation(Xtest3)Deviation(Xtest2)Deviation(Xtest1)Sepal_length(X4)Sepal_width(X3)Petal_length(X2)Petal_width(X1)Species
Memory Based Classification based on training set
Now using the X1(test) , X2(test) and X3(test) vectors to compute which is theclosest approximation to the input values
-
8/9/2019 Working of Neural Networks
11/12
141052
w14w13w12w11Final Weights
14.478810.22125.113811.520144.470052.682031.4304170.1791230.065138194Setosa
10.00877.539163.683391.341343.336902.486321.0468720.131788.2305030162Setosa
6.671865.052842.636521.210481.811391.317370.5351830.082576.8785138162Setosa
4.860473.73542.101331.128141.554281.141920.4440800.032362.9694432132Setosa
3.306182.593541.657251.096421.041550.672670.3471860.043279.6694936141Setosa
2.264631.920871.310071.053030.703200.550330.1528690.031191.3164831162Setosa
1.561421.370541.15721.022450.561420.370540.1572020.022134.7824636102Setosa
11110000995033142Setosa
w14w13w12w11w14w13w12w11Yk(X4)(X3)(X2)(X1)Species_Name
0.0001Learning rate(n)=
Neuron 1 (Training Set)
Hebbian learning which computes changes in weights wjk by multiplying thepresynaptic weight Xj with post synaptic weight Yk and learning rate (n)
-
8/9/2019 Working of Neural Networks
12/12
This graph proves hebbian learning because it shows that there is increaseIn weight when the same presynaptic weight is applied to the same neuronTo produce synchronous post synaptic weight