Результаты обучения полной нейронной сети с N=3,4,5,6 нейронами на матрице


при различных активаторных фунциях и методах минимизации

3 Нейрона:


fA=arctan(x)
Meth=Coord
Nerr=6, S= 1.21
fA=1/(1+exp(-x))
Meth=Coord
Nerr=6, S= 1.17
fA=RELU
Meth=Coord
Nerr=1, S= 0.6
fA=RELU_1
Meth=Coord
Nerr=4, S= 1.14

fA=arctan(x)
Meth=Grad
Nerr=6, S= 1.21
fA=1/(1+exp(-x))
Meth=Grad
Nerr=6, S= 1.17
fA=RELU
Meth=Grad
Nerr=2, S= 0.56
fA=RELU_1
Meth=Grad
Nerr=4, S= 1.19

fA=arctan(x)
Meth=Co+Gr
Nerr=6, S=1.21
fA=1/(1+exp(-x))
Meth=Co+Gr
Nerr=6, S=1.17
fA=RELU
Meth=Co+Gr
Nerr=1, S=0.6
fA=RELU_1
Meth=Co+Gr
Nerr=4, S=1.1

4 Нейрона:


fA=arctan(x)
Meth=Coord
Nerr=0, S=1.08
fA=1/(1+exp(-x))
Meth=Coord
Nerr=3, S=1.06
fA=RELU
Meth=Coord
Nerr=1, S=0.18
fA=RELU_1
Meth=Coord
Nerr=0, S=0.7

fA=arctan(x)
Meth=Grad
Nerr=0, S=1.08
fA=1/(1+exp(-x))
Meth=Grad
Nerr=3, S=1.06
fA=RELU
Meth=Grad
Nerr=1, S=0.18
fA=RELU_1
Meth=Grad
Nerr=0, S=0.7

fA=arctan(x)
Meth=Co+Gr
Nerr=0, S=1.08
fA=1/(1+exp(-x))
Meth=Co+Gr
Nerr=3, S=1.06
fA=RELU
Meth=Co+Gr
Nerr=1, S=0.18
fA=RELU_1
Meth=Co+Gr
Nerr=0, S=0.71

5 Нейронов:


fA=arctan(x)
Meth=Coord
Nerr=0, S=0.93
fA=1/(1+exp(-x))
Meth=Coord
Nerr=0, S=0.92
fA=RELU
Meth=Coord
Nerr=0, S=???
fA=RELU_1
Meth=Coord
Nerr=0, S=0.73

fA=arctan(x)
Meth=Grad
Nerr=0, S=0.93
fA=1/(1+exp(-x))
Meth=Grad
Nerr=1, S=0.88
fA=RELU
Meth=Grad
Nerr=0, S=???
fA=RELU_1
Meth=Grad
Nerr=0, S=0.68

fA=arctan(x)
Meth=Co+Gr
Nerr=0, S=0.93
fA=1/(1+exp(-x))
Meth=Co+Gr
Nerr=1, S=0.88
fA=RELU
Meth=Co+Gr
Nerr=0, S=0.07
fA=RELU_1
Meth=Co+Gr
Nerr=0, S=0.7

6 Нейронов:


fA=arctan(x)
Meth=Coord
Nerr=0, S=0.83
fA=1/(1+exp(-x))
Meth=Coord
Nerr=0, S=0.77
fA=RELU
Meth=Coord
Nerr=0, S=???
fA=RELU_1
Meth=Coord
Nerr=0, S=0.67

fA=arctan(x)
Meth=Grad
Nerr=0, S=0.86
fA=1/(1+exp(-x))
Meth=Grad
Nerr=0, S=0.78
fA=RELU
Meth=Grad
Nerr=0, S=???
fA=RELU_1
Meth=Grad
Nerr=0, S=0.68

fA=arctan(x)
Meth=Co+Gr
Nerr=0, S=0.86
fA=1/(1+exp(-x))
Meth=Co+Gr
Nerr=0, S=0.8
fA=RELU
Meth=Co+Gr
Nerr=0, S=???
fA=RELU_1
Meth=Co+Gr
Nerr=0, S=0.68