Optimization of multi-layer artificial neural networks using delta values of hidden layers

Wagarachchi, N. Mihirini; Karunananda, AS

doi:10.1109/ccmb.2013.6609169

Cited by 12 publications

(11 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, the nodes number of the second hidden layer depends on the number of input nodes. So in this paper, we set the nodes number of the second hidden layer as (2m + 1)/2 (m is the nodes number of the input layer) according to Hecht-Nielsen's method [25], which is simple and effective.…”

Section: The Methods Of Selecting the First Hidden Layer Nodes Number mentioning

confidence: 99%

“…Genetic algorithm was proposed by J.H. Holland in 1975 inspired by species evolution [29]. Genetic algorithm simulates the evolution process of biology in nature, which means that survival of the fittest in natural selection.…”

Section: Improved Genetic Algorithm To Train Learning Rate and Momentmentioning

confidence: 99%

“…We suppose the accuracy of solution (training results) are three decimal. Binary representation of chromosome should satisfy Len = L/A, where Len is the length of binary representation of solution; L is the interval length of solution domain; A is the accuracy of solution [29]. In order to ensure accuracy and make the binary representation as short as possible, we set the solution domain of Lr in [0.05, 0.1] and the solution domain of M c in [0.6, 1.0] by experience and many times experiments.…”

Section: Improved Genetic Algorithm To Train Learning Rate and Momentmentioning

confidence: 99%

“…Besides, the nodes number of hidden layer is crucial for network training accuracy. However, existing neural network models often set the nodes number of hidden layer depend on artificial or mathematical formulas [14][15][16], which is too subjective and cannot adjust dynamically. In addition, learning rate and momentum factor are two important parameters for training accuracy of neural network.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Improved Artificial Neural Network Based on Intelligent Optimization Algorithm

Xu¹,

He²

2018

NNW

View full text Add to dashboard Cite

Neural network based on back-propagation (BP) algorithm is a widely used prediction model. However, the nodes number of the first hidden layer, the learning rate and momentum factor are usually determined manually, which affects the forecast accuracy of network. Therefore, in this paper, to improve the forecast accuracy, firstly, the nodes number of the first hidden layer is selected adaptively based on minimizing mean square error (MSE). Secondly, improved genetic algorithm (GA) is proposed to train the learning rate and momentum factor dynamically, which includes multi-point crossover and single point mutation. Thirdly, we construct a new neural network model based on the adaptively selected nodes number of the first hidden layer, the dynamically selected learning rate and momentum factor, which is called HN-GA-BP neural network model. Finally, the proposed neural network model is used to forecast the carbon dioxide contents in China for fifty years. Experimental results demonstrate the effectiveness of the proposed HN-GA-BP neural network model.

show abstract

Section: The Methods Of Selecting the First Hidden Layer Nodes Number mentioning

confidence: 99%

Section: Improved Genetic Algorithm To Train Learning Rate and Momentmentioning

confidence: 99%

Section: Improved Genetic Algorithm To Train Learning Rate and Momentmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Improved Artificial Neural Network Based on Intelligent Optimization Algorithm

Xu¹,

He²

2018

NNW

View full text Add to dashboard Cite

show abstract

“…Sometimes the network matches the data narrowly and loses its generalization ability over the test data which give rise to overtraining. In [37], author gives information regarding the competitive learning approach of finding hidden nodes for artificial neural network. The advantage of this proposed method is that it is not approximately calculating number of hidden nodes but based on similarity between input data.…”

Section: Literature Surveymentioning

confidence: 99%

Optimization of BPNN parameters using PSO for EEG signals

Kshirsagar¹,

Akojwar²

2017

Proceedings of the International Conference on Communication and Signal Processing 2016 (ICCASP 2016)

View full text Add to dashboard Cite

Epilepsy is a neurological disorder characterized by the existence of recurring seizures. Like many other neurological disorders, epilepsy can be appraisal by the electroencephalogram (EEG). The EEG signal is highly non-linear and non-stationary and consist of lot of data including significant data and artifact and hence, it is practically arduous to characterize and interpret it. However, it is a well-established clinical technique with low associated costs for detection of various neurological disorders. In this work, we propose a methodology for the automatic detection of normal, epilepsy and brain death from recorded EEG signals collected from clinic. Discrete wavelet transform is applied for feature extraction. Back Propagation neural network optimized by particle swarm optimization is used for classification of neurological disorders. Simple BPNN has several drawbacks which mainly include large time duration during EEG signal classification. This drawback is removed by PSO. In this paper, the proposed method used to detect the number of neurons in hidden layer of BPNN using optimization technique of PSO. Once the numbers of neurons in hidden layers are detected, optimum value for initial weights and biases for BPNN estimated which is further used for classification and sortilege of various neurological disorders. So that time duration decreases and accuracy increases. EEG signals are recorded for 30 minutes of three different patients that are epileptic, normal and brain death used to rehearse and test the proposed algorithm. A signal used to test is integrated signal by taking mean of 16 channels. By applying techniques to signals of epilepsy or normal or brain death patient which are known to us we find the more accurate results with less number of iteration and time .

show abstract

End‐to‐end inverse modeling and optimization for filters based on artificial neural network

Huang

Liang

Guan

et al. 2022

Int J RF Mic Comp-Aid Eng

View full text Add to dashboard Cite

In this letter, we propose an end-to-end inverse modeling and optimization method for microwave filter designs based on the data-augmentation learning strategy. Because of the non-uniqueness of solutions, it is difficult to achieve good convergence with artificial neural networks for inverse designs when the parameter space is very large. We prove that the accuracy of inverse predictions can be significantly improved using the network's self-generated data and the optimization can be greatly accelerated with the help of the inverse network. The predicted structural parameters can be used as initial values for optimization, which reduces the number of iterations and avoids falling into local optima. This method is applied to designs of fourth-order interdigital cavity filters. The measurement and simulation agree well.

show abstract

Optimization of multi-layer artificial neural networks using delta values of hidden layers

Cited by 12 publications

References 5 publications

Improved Artificial Neural Network Based on Intelligent Optimization Algorithm

Improved Artificial Neural Network Based on Intelligent Optimization Algorithm

Optimization of BPNN parameters using PSO for EEG signals

End‐to‐end inverse modeling and optimization for filters based on artificial neural network

Contact Info

Product

Resources

About