[1]张 胜,王 蔚,刘红星,等.基于构造性设计和GA的ANN建模方法[J].南京师大学报(自然科学版),2008,31(04):61-65.
 Zhang Sheng,Wang Wei,Liu Hongxing,et al.An ANN Model Based on Constructive Algorithm and GA[J].Journal of Nanjing Normal University(Natural Science Edition),2008,31(04):61-65.
点击复制

基于构造性设计和GA的ANN建模方法()
分享到:

《南京师大学报(自然科学版)》[ISSN:1001-4616/CN:32-1239/N]

卷:
第31卷
期数:
2008年04期
页码:
61-65
栏目:
物理学
出版日期:
2008-12-30

文章信息/Info

Title:
An ANN Model Based on Constructive Algorithm and GA
作者:
张 胜1 王  蔚2 刘红星3 余水宝1
( 1. 浙江师范大学数理与信息工程学院, 浙江金华321004)
( 2. 南京师范大学教育科学学院, 江苏南京210097)
( 3. 南京大学电子科学与工程系, 江苏南京210093)
Author(s):
Zhang Sheng1Wang Wei2Liu Hongxing3Yu Shuibao1
1.College of Mathematics,Physics and Information Engineering,Zhejiang Normal University,Jinhua 321004,China
关键词:
人工神经网络 构造性设计 GA 激活函数类型优化
Keywords:
ANN constructive a lgo rithm GA activa tion functions tapes
分类号:
TP183
摘要:
强调了激活函数在ANN设计中的重要性,提出一种基于构造性设计及GA的网络结构及神经元激活函数类型自动优化的ANN模型(constructived and GA based activation function,简称为CGBAF),并给出其一般形式和算法.本模型用于多层前向神经网络时,其网络结构及激活函数类型可自动优化,进而可大大提高ANN的泛化能力.通过例子验证了本方法的有效性,并进行了分析.
Abstract:
The importance o f the activa tion functions in ANN is emphasized. A new ANN m ode ling m ethod is proposed based on construc tive a lgor ithm and GA. Th ism e thod can be used to realize the autom a tic op tim ization of the ne t structure and the types o f the activ ation functions. As a result, the ANN’ s gene ra liza tion capability is g reatly improved. Th is im provement is ver ified exper imenta lly.

参考文献/References:

[ 1]  Zhang Sheng. Survey ing them ethods of improv ing ANN gene ra liza tion capability[ C ] / / Pro ceedings o f the Second International Conference onM achine Learning and Cybernetics, X i‘ an, 2003: 1 259-1 263.
[ 2]  张建勋, 贺京同, 王维, 等. 人工神经网络对时间增长序列预测能力分析[ J]. 预测, 1999, 5: 60-63.
[ 3]  吴佑寿, 赵明生. 激活函数可调的神经元模型及其有监督学习的应用[ J]. 中国科学( E 辑), 2001, 31( 3): 263-272.
[ 4]  W ang Z O, Zhu T. An e fficien t learn ing a lgo rithm for im prov ing generalization perform ance o f radial basis func tion neural netwo rks[ J]. Neura lNe tw orks, 2000, 13( 4 /5): 545-553.
[ 5]  鲁子弈, 杨绿溪, 无球, 等. 提高前馈神经网络泛化能力的新算法[ J]. 电路与系统学报, 1997, 2( 4) : 7-12.
[ 6]  彭汉川, 甘强, 韦钰. 提高前馈神经网络推广能力的若干实际方法[ J] . 电子学报, 1998, 26( 4): 116-119.
[ 7]  G iles C L. Constructive learning of recurrent neura l netwo rks: prob lem sw ith recurrent cascade corre lation and a sim ple so lution[J]. IEEE T ransactions on Neura lNetw orks, 1995, 6( 4): 829-836.
[ 8]  Kw ok T Y, Y eung D Y. Construc tive a lgor ithms for structure learn ing in feed fo rw ard neura l netwo rks fo r regress ion prob lem s[ J] . IEEE T ransactions on N eural Netwo rks, 1997, 8( 3): 630-645.[ 9]  T readgo ld N K, G edeon T D. Exploring constructive cascade netwo rks[ J]. IEEE T ransactions on Neura l Netwo rks, 1999,10( 6): 1 335-1 350.
[ 10]  Cybenco G. Approx im ations by supe rpo sitions of a sigmo id function[ J]. M a th Cont S ignal Syst, 1989, 2: 303-314.
[ 11] H ornik K. Approx im ation capabilities of mu ltilayer feedfo rw ard netw orks[ J]. Neural Netwo rks, 1991, 4: 251-257.
[ 12]  M acKay. Bayesian inte rpo lation[ J]. N eural Computation, 1992, 4( 3) : 415-447.
[ 13]  Foresee F D, H aganM T. Gauss-New ton approx im ation to Bayesian regulariza tion[ C ] / / Proceed ing s o f the 1997 International Jo int Conference on Neural Netwo rks. N ew York, 1997: 1 930-1 935.
[ 14] M o llerM F. A sca led conjugate g radien t algor ithm for fast supe rv ised learn ing [ J] . Neura l Netwo rks, 1993, 6: 525-533.

备注/Memo

备注/Memo:
基金项目: 国家自然科学基金( 59905011)、浙江省自然科学基金(Y207738)资助项目.
通讯联系人: 张  胜, 博士, 副教授, 研究方向: 电子建模. E-mail:zs@ zjnu. cn
更新日期/Last Update: 2013-05-05