Artificial Speciation andAutomatic Modularisation
Artificial Speciation and Automatic Modularisation
Neural Network Ensemble A group of neural networks is used to solve aproblem; each perhaps concentrating on part of thedataset The final output of the ensemble is determined bycombining the output of individual NN using:- Majority Voting- Simple Averaging- Weighted average· Evolutionary NNE: Ensemble = whole population/a sub-population formed by clustering
Neural Network Ensemble • A group of neural networks is used to solve a problem; each perhaps concentrating on part of the dataset • The final output of the ensemble is determined by combining the output of individual NN using: − Majority Voting − Simple Averaging − Weighted average • Evolutionary NNE: Ensemble = whole population/ a sub-population formed by clustering
Distance Measure? Distance between two NN, D(p,q), ismeasured by the cross-entropy:D(P,9)=(P, og +4 og2qjpii=1. Low D(p,q) means p and q are very similar
Distance Measure • Distance between two NN, D(p,q), is measured by the cross-entropy: • Low D(p,q) means p and q are very similar ( log log ) 2 1 ( , ) 1 j j j n j j j j p q q q p D p q = p + =
Fitness Sharing Speciation at the phenotypic (behavior) levelRaw fitness:1tf raw,pMSEpShared fitnessraw.phainZs(D(p,q,)j-1 The sharing function, S(D(p,g), can be linear- Gaussian distribution in this case?
Fitness Sharing • Speciation at the phenotypic (behavior) level • Raw fitness: • Shared fitness: • The sharing function, S(D(p,q)), can be linear − Gaussian distribution in this case? p raw p MSE f 1 , = = = n j j raw p shared p s D p q f f 1 , , ( ( , ))
Negative Correlation Learning Error of NN i for the nth training patternNE, =22(F(m)-d(m),+ap,(n)Nn=l. The penalty termp,(n)=(F(n)-F(n)(F,(n)-F(n)j+i
Negative Correlation Learning • Error of NN i for the nth training pattern • The penalty term ( ) 1 ( ( ) ( )) 2 1 1 1 1 2 p n N F n d n N E i N n N n i i = = = − + p (n) (F (n) F(n)) (F (n) F(n)) j i i = i − j −