|
OALib Journal期刊
ISSN: 2333-9721
费用：99美元
|
|
|
|
基于构造性设计和ga的ann建模方法
, PP. 61-65
Keywords: 人工神经网络,构造性设计,ga,激活函数类型优化
Abstract:
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(constructivedandgabasedactivationfunction,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.
| [1] | 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.
|
| [2] | wangzo,zhut.anefficientlearningalgorithmforimprovinggeneralizationperformanceofradialbasisfunctionneuralnetworks[j].neuralnetworks,2000,13(4/5):545-553.
|
| [3] | 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.
|
| [4] | gilescl.constructivelearningofrecurrentneuralnetworks:problemswithrecurrentcascadecorrelationandasimplesolution[j].ieeetransactionsonneuralnetworks,1995,6(4):829-836.
|
| [5] | kwokty,yeungdy.constructivealgorithmsforstructurelearninginfeedforwardneuralnetworksforregressionproblems[j].ieeetransactionsonneuralnetworks,1997,8(3):630-645.[9]treadgoldnk,gedeontd.exploringconstructivecascadenetworks[j].ieeetransactionsonneuralnetworks,1999,10(6):1335-1350.
|
| [6] | zhangsheng.surveyingthemethodsofimprovinganngeneralizationcapability[c]//proceedingsofthesecondinternationalconferenceonmachinelearningandcybernetics,xi‘an,2003:1259-1263.
|
| [7] | 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.
|
| [8] | 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.
|
| [9] | cybencog.approximationsbysuperpositionsofasigmoidfunction[j].mathcontsignalsyst,1989,2:303-314.
|
| [10] | hornikk.approximationcapabilitiesofmultilayerfeedforwardnetworks[j].neuralnetworks,1991,4:251-257.
|
| [11] | mackay.bayesianinterpolation[j].neuralcomputation,1992,4(3):415-447.
|
| [12] | foreseefd,haganmt.gauss-newtonapproximationtobayesianregularization[c]//proceedingsofthe1997internationaljointconferenceonneuralnetworks.newyork,1997:1930-1935.
|
| [13] | mollermf.ascaledconjugategradientalgorithmforfastsupervisedlearning[j].neuralnetworks,1993,6:525-533.
|
Full-Text
|
|
Contact Us
service@oalib.com QQ:3279437679 
WhatsApp +8615387084133
|
|
