ѧ
JOURNAL OF SOFTWARE
1999 10 12  Vol.10 No.12 1999



ŴԺķǼල෽
  ϳ ߺ
ժҪһµĻŴԺģART(adaptive resonance theory)ķǼල෽.,ԭеѵģARTзǼලѵȻ,ŴΪģARTӸ߽ڵѵ㣬ٶģARTмලѵ.ַѵڽµARTϵѧϰ⣬ARTϵķܣչӦ÷Χ.
ؼʡ磬Ŵ㷨ART(adaptive resonance theory)Ǽල.
ͼšTP18
The Unsupervised Classification Using Evolutionary Strategies and Neural Networks
LI Ming YAN Chao-hua LIU Gao-hang
(Department of Applied Engineering Nanchang Institute of Aeronautical Technology Nanchang 330034)
Abstract A new unsupervised classification method using evolutionary strategies and fuzzy ART (adaptive resonance theory) neural networks is proposed in this paper. First, fuzzy ART neural networks is trained by original input samples under unsupervised way. Then evolutionary strategies is used to generate new training samples near the clusters boundaries of neural networks. Therefore the weights of fuzzy ART neural networks can be revised and refined by training those new generated samples under supervised way. The proposed method resolves the training problem for ART serial neural networks when there are only less training samples available. Consequently, it enhances the performance of ART serial neural networks and extends their application.
Key wordsNeural network, genetic algorithms, ART(adaptive resonance theory), unsupervised classification.
ڶķǼල෽УCarpenter14ݵ˽ͷչĻӦг(adaptive resonance theoryART)һϵЧķ֮һ.ARTϵѧϰУռ䲻ͬλõԽ࣬Խȷ.ʵӦֻܻýٵ㣬磬ģȱָƻʶӦУѻøָľģȱ.,άCADƵķӦҲƵ.ٵѵARTܽϲԴ߽ڵ㣬ARTķ಻ɿ.
Hwang5ݷ֣ͨԶ֪ԤѵЧ߷.ȱٶѵ£Thomas6ݲƵķΪART1ѵַŴ㷨ɵϺɶϵͳɵӣ̽ӵ״.ThomasķУȷһЩŴȺ壬ȻͨѡŴӵе丸ĸеART1ѧϰı߽λ.Ŵ㷨ʱThomasõART1ѧϰı߽總㣬ȻЩµһѵART1磬˷ȷ.
ThomasֻܶԶֵŴ㣬˶Զֱֵӱ뷽ʽԼӦĶԶֵĽѡŴ.ֻĽֵֻܶART1磬ӦڿɽʵֵART2ģART.ңThomasֻǼ򵥵µѵART1磬µѵART1״ȷֱӹϵѵART1һЩı߽總ķ.
һµĻŴԺģARTķǼල෽3.1ԭеѵģARTзǼලѵģARTȨʸ.2Ŵ(evolution strategies)ΪģARTµĴѧϰ߽總.3ȷµ壬ͬԭеѵһģARTмලѵӦصģARTľԲʹģART.
µķǼල෽ͨŴԵŴ̽ѵڽµARTϵѵ뽨⣬淢չThomasһṩʵֵӷʹʺ͵ARTϵ磻мලѵϢARTϵķ.
1 ģARTĳʼ
ķǼල෽ĵ1ԭеѵģARTзǼලѵģARTȨʸʵַǼලĳʼ.
ĲõĽмලͷǼලѵģARTͼ1ʾͬʱмලͷǼලѵ.ڽзǼලѵʱԤ벻ڣһ׼ģART磬Ϊͼ1еARTa.ڽмලѵʱһ򻯵ģARTMA3磬ԤֱӸ壬ȥģARTMAPеһ׼ģARTARTb3.

ͼ1ĲõģARTṹ
׼ģARTÿһԪһ壬ÿռ״λȨʸÿռпһ2,4.ʸΪ룺ΪMάʸ.ɶȨʸɵȨʸȫͬģΪ辯Ϊ.ӦjԪΪԽΪһRj.εԶԭĶΪԭĶΪ.j1ѡѧϰʱȨʸΪҲ,ĳRjһMάĵ.εĳߴ綨Ϊڿѧϰ£RjչΪҲǣR(new)jչܰԭR(old)jMά.Ϊ˼öάʸ(={a1,a2},M=2)ͼʾĳμν.չͼ2ʾͼжάεĳߴڳεĳȺͿ֮.ɲοף2,4ݿɵãչ˵ĳγߴ㾯ʱjͿ㾯.ھѡУѡԭεĳߴչĳγߴ.,ģARTʵַǼලľͷļνΪ

ͼ2εչʾͼ
(1) ʸΨһһѽĳRjУ˳һܹͨԲұѡ.
(2) ʸڶѽĳУСγߴ彫ѡ.
(3) ʸκѽĳУߴ㾯ĳμΪʽRjѡ.
(1)
(4) ûԤΪʸһ.
2 Ŵ߽
Ŵԣ79봮ʾһŻ⣬ɲͬ봮ŴȺ.ŴԶ봮Ŵµ봮ʹµ봮иߵӦ.Ĳ(,)Ŵ.
Ŵ봮Ϊ2MάʸaiΪʸĵiiΪӦĲ.봮p()䲹()1ѵõģART磬ȡ׼弯(Ҳǳμ).
(2)
봮p()ӦΪ
(3)
ЦС.KLʽ
(4)
(5)
,ӦĳRKͳRL㳬ξ׼ʸΣ봮p()Ӧfitness()ʸľ֮봮p()Ӧfitness()ԽʸԽڳRKͳRLмλãҲ˵ʸԽı߽總.
(,)ŴһľʱԵõڲ߽ͬ總ʸ.ЩµʸĲ̣Ϊÿµʸ()ı߽總ʹЩµʸһڽܵмලģARTѵ.
3 мලģART׼ȷ
ThomasֻǼ򵥵µѵART1磬µѵӰART1ѵѵART1һЩı߽總ķ.ͬ򵥵µѵģART,ҲƵ.ͼ3ʾͼ4ߴͬĳΣRARBRCRDСڵʾͨŴõλ4мӵ.µѵӰģARTѵ.¿ʼֽܵĳչ볬RBRDмӵ㣬򳬷RB򳬷RDչ.ڳRBĳߴչýϴ󣬵볬RBRBмӵʱеĶΪRBӦ壬ֻΪRBӦ壬RBRBӦѧϰͼ4ʾ.

ͼ3ѱѵģARTĳμӵʾͼ

ͼ4ӵһѧϰ
볬RBRDмӵ㣬볬RARBмӵ㣬볬RBRCмӵ㣬RBRBӦѧϰͼ5ʾ.ݾ׼չγߴʱٴѵģARTԳRBRBмӵķΪ.

ͼ5ӵһѧϰ
ĶԵ1еõģARTмලѵ.мලѵģARTԳŴԵõӵϢӦԲģARTʶ.
мලģARTģARTMAP磬Ĺԭμף2,3.ŴԵ̿жÿӵ㴦ı߽㸽.ӵ㴦ı߽㸽
(6)
ӵĸжϹΪ
(7)
Ȼ,ЩѱӵԵ1о˳ʼѵģARTٴεмලѵ.ͼ1ʾĳӵʱͬʱ.ͼ1ART\-aԤaͬARTaб׼ѵѧϰ.Ԥaͬͼ1еMAPĴźXabӦظıARTaľԲѣʹARTaԤͬ.ӵжмලѵʹģARTѧϰģARTµԽи׼ȷķ.
4 
Ϊ˼ʾŴԺģARTķǼල෽ʵĽֻжάռĳʶ.ռɳεĳȺͿɣһ೤εĳȴڿȣһ೤εĿȴڳ.
ΪؿŴҳģARTĳչֻ˽ٵĳʼ,ʹ÷ǼලѧϰģARTֻRARB.ͼ6ʾRAǿWidthڳLengthĳ壬RBǳLengthڿWidthĳ壬ռǶάģRARBʵϾͼ6ʾ.ģARTڷǼලѵʼʱԲΪ=0.35򳬷RARBԼչĳεĳߴСڻM(1-)=2(1-0.35)=1.3.

ͼ6ԭʼѵõģARTĳ
Ŵ1 000Σõ26ͼ7еʾǴڳRARBмλ.ЩģARTмලѧϰȶ󣬳RARBֱչΪͼ7е߷αʾĳΣʱľԲΪ=0.59.ͼ6ͼ7ʾٵ£ŴԺģARTķǼල෽õ״Ϊ˷ǼලĿɿ.

ͼ7ලѵõģARTĳ
ĻŴԺģARTķǼල෽ʶ.ռ10Ƶ5Ҫ(ߡߡߡߺǻ)λþռ乲25ά.ģARTķǼල෽ʶ𷽷μף10.ʶ𷽷뱾ĻŴԺģARTķǼල෽ʶԱȽ1.ʶй50࣬5 000ÿ100.ڽַǼලѵʱ,ÿֻѡ10ѧϰ500.1ַѵõģART5 000ȷʶʣӦľԲ.ڻŴԺģARTķǼල෽УŴԽ5 000Ẉ̂40мලѵ.
1 ±ĵķ෽ģART෽ʶ

ȷʶ(%)Բ
ģARTķǼල෽93.60.84
ŴԺģARTķǼල෽95.70.81

˻ŴԺģARTķǼල෽ͨŴԵŴ̺ͲARTмලǼලѵµARTϵѵ뽨.ṩӶֵֵķмලѵϢARTϵķ.
*оõʡȻѧ.
߼飺,1965,ʿоԱ,ҪоΪͼģʽʶ磬Ŵ㷨.
ϳ,1956,ڣҪоΪԼͼ.
ߺ,1946,ڣҪоΪԼģʽʶ.
ͨѶϵ:ϲ330034ϲչҵѧԺӦùϵ
ߵλϲչҵѧԺӦùϵ ϲ 330034
ο
1 Carpenter G A, Grossberg S. A massively parallel architecture for a self-organizing neural pattern recognition machine. Computer Vision, Graphics, and Image Processing, 1987,37(1):54115
2 Carpenter G A, Grossberg S, Rosen D B. Fuzzy ART: fast stable learning and categorization of analog patterns by an adaptive resonance system. Neural Networks, 1991,4(6):759771
3 Carpenter G A, Grossberg S, Markuzon N et al. Fuzzy ARTMAP: a neural network architecture for incremental supervised learning of analog multidimensional supervised maps. IEEE Transactions on Neural Networks, 1992,3(5):698613
4 Huang J, Georgiopoulos M, Heileman G L. Fuzzy ART properties. Neural Networks, 1995,8(2):202213
5 Hwang J N, Choi J J et al. Query leaning based on boundary search and gradient computation of trained multiplayer perceptions. In: Proceedings of the International Joint Conference on Neural Networks. San Diego: IEEE Press, 1990. 433442
6 Thomas P C. Genetic Algorithms as a tool for the analysis of adaptive resonance theory network training sets. In: Proceedings of the International Workshop on Combinations of Genetic Algorithms and Neural Networks. Baltimore: IEEE Press, 1992. 184200
7 David B F. An introduction to simulated evolutionary optimization. IEEE Transactions on Neural Networks, 1994,5(1):314
8 Back T, Schwefel H P. An overview of evolutionary algorithms for parameter optimization. Evolutionary Computation, 1993,1(1):124
9 Yao X. A review of evolutionary artificial neural networks. International Journal of Intelligent Systems, 1993,8(4):539567
10 Paul S W, Li M. Personal identification by palm prints recognition. In: Proceedings of the 10th International FLAIRS Conference. Daytona Beach: IEEE Press, 1997. 12111218
1998-08-12յԭ,1998-12-28յ޸ĸ
