Research Article  Open Access
FractionalOrder Deep Backpropagation Neural Network
Abstract
In recent years, the research of artificial neural networks based on fractional calculus has attracted much attention. In this paper, we proposed a fractionalorder deep backpropagation (BP) neural network model with regularization. The proposed network was optimized by the fractional gradient descent method with Caputo derivative. We also illustrated the necessary conditions for the convergence of the proposed network. The influence of regularization on the convergence was analyzed with the fractionalorder variational method. The experiments have been performed on the MNIST dataset to demonstrate that the proposed network was deterministically convergent and can effectively avoid overfitting.
1. Introduction
It is well known that artificial neural networks (ANNs) are the abstraction, simplification, and simulation of the human brains and reflect the basic characteristics of the human brains [1]. In recent years, great progress has been made in the research of deep neural networks. Due to the powerful ability of complex nonlinear mapping, many practical problems have been successfully solved by ANNs in the fields of pattern recognition, intelligent robot, automatic control, prediction, biology, medicine, economics, and other fields [2, 3]. BP neural network is one of the most basic and typical multilayer forward neural networks, which are trained by backpropagation (BP) algorithm. BP, which is an efficient way for optimization of ANNs, was firstly introduced by Werbos in 1974. Then, Rumelhart and McCelland et al. implemented the BP algorithm in detail in 1987 and applied it to the multilayer network version of Minsky [4–6].
The fractional calculus has a history as long as the integral order calculus. In the past three hundred years, the theory of fractional calculus has made great progresses [7–11]. Its basics are differentiation and integration of arbitrary fractional order. Nowadays, fractional calculus is widely used in diffusion processes [12–14], viscoelasticity theory [15], automation control [16–18], signal processing [19–21], image processing [22–25], medical imaging [26–28], neural networks [29–37], and many other fields. Due to the longterm memory, nonlocality, and weak singularity characteristics [29–37], fractional calculus has been successfully applied to ANNs. For instance, Boroomand constructed the Hopfield neural networks based on fractional calculus [37]. Kaslik analyzed the stability of Hopfield neural networks [30]. Pu proposed a fractional steepest descent approach and offered a detailed analysis of its learning conditions, stability, and convergence [38]. Wang applied the fractional steepest descent algorithm to train BP neural networks and proved the monotonicity and convergence of a threelayer example [33]. However, there are three limitations in the proposed fractionalorder BP neural network models in [33]. First, the neural network in [33] just had 3 layers, which was actually a shadow network and was not proper to demonstrate its potential for deep learning. Second, the fractional order of this model was restricted to without reasonable analysis. Third, the loss function did not contain the regularization term, which is an efficient way to avoid overfitting, especially when the training set has small scalar. Overfitting means that the model has high prediction accuracy on training set but has the low prediction accuracy on testing set. This makes the generalization ability of the model poor, and the application value is greatly reduced.
In this paper, we proposed a deep fractionalorder BP neural network with regularization term, and the fractionalorder could be any positive real number. With the fractionalorder variational method, the influence of regularization on the convergence of the proposed model was exploited. The performance of the proposed model was evaluated on the MINST dataset.
The structure of the paper is as follows: in Section 2, the definitions and simple properties of fractional calculus are introduced. In Section 3, the proposed fractionalorder multilayer BP neural networks are given in detail. In Section 4, the necessary conditions and the influence of regularization for the convergence of the proposed BP algorithm are stated. In Section 5, experimental results are presented to illustrate the effectiveness of our model. Finally, the paper is concluded in Section 6.
2. Background Theory for Fractional Calculus
In this section, the basic knowledge of fractional calculus is introduced, including the definitions and several simple properties used in this paper.
Different from integer calculus, fractional derivative does not have a unified temporal definition expression up to now. The commonly used definitions of fractional derivative are GrünwaldLetnikov (GL), RiemannLiouville (RL), and Caputo derivatives [7–11].
The following is the GL definition of fractional derivative:where denotes the fractional differential operator based on GL definition, denotes a differintegrable function, is the fractional order, is the domain of , is the Gamma function, and [] is the rounding function.
The RL definition of fractional derivative is as follows: where denotes the fractional differential operator based on GL definition; . Moreover, the GL fractional derivative can be deduced from the definition of the RL fractional derivative.
The Caputo definition of fractional derivative is as follows:where is the fractional differential operator based on Caputo definition, .
Fractional calculus is more difficult to compute than integer calculus. Several mathematical properties used in this paper are given here. The fractional differential of a linear combination of differintegral functions is as follows:where and are differintegral functions and and are constants.
The fractional differential of constant function , ( is a constant) is different under different definitions:
For the GL definition,
For the RL definition,
And for the Caputo definition
According to (6), (7) and (8), we can know that for the GL and RL definition, the fractional differential of constant function is not equal to 0. Only with the Caputo definition, the fractional differential of constant function equals to 0, which is consistent to the integerorder calculus. Therefore, the Caputo definition is widely used in solving engineering problems and it was employed to calculate the fractionalorder derivative in this paper. The fractional differential of function , is as follows:
3. Algorithm Description
3.1. FractionalOrder Deep BP Neural Networks
In this section, we introduce the fractionalorder deep BP neural network with layers. , , is the number of neurons for the th layer. denotes the weight matrix connecting the th layer and the th layer. denotes the corresponding activation function for the th layer. and are the input and the corresponding ideal output of the th sample and the training sample set is . denotes the total inputs of th layer. If neurons in the th layer are not connected to any neurons in previous layer, these neurons are called external outputs of the th layer, denoted as . On the contrary, if neurons in the th layer are connected to every neuron in previous layer, these neurons are called internal outputs of th layer, denoted as . denotes the total outputs of th layer. The forward computing of the fractionalorder deep BP neural networks is as follows:
Particularly, external outputs can exist in any layer except the last one. With the square error function, the error corresponding to th sample can be denoted as:where denotes the th element of , denotes the th element of .
The total error of the neural networks is defined as
In order to minimize the total error of the fractionalorder deep BP neural network, the weights are updated by the fractional gradient descent method with Caputo derivative. Let . The backpropagation of fractionalorder deep BP neural networks can be derived with the following steps.
Firstly, we define that
According to (13), we can know that
Then the relationship between and can be given by
Then, according to the chain rule and (17), we have
The updating formula iswhere denotes the th iteration and is the learning rate.
3.2. Fractional Deep BP Neural Networks with Regularization
Fractionalorder BP neural network can be overfitted easily when the training set has small scalar. regularization is a useful way to avoid models to be overfitted without modifying the architecture of network. Therefore, by introducing the regularization term into the total error, the modified error function can be presented aswhere denotes the sum of squares of all weights and denotes the regularization parameter.
By introducing (18), we have
The updating formula iswhere denotes the th iteration and is the learning rate.
4. Convergence Analysis
In this section, the convergence of the proposed fractionalorder BP neural network is analyzed. According to previous studies [39–42], there are four necessary conditions for the convergence of BP neural networks:
(1) The activation functions are bounded and infinitely differentiable on R and all of their corresponding derivatives are also continuous and bounded on . This condition can be easily satisfied because the most common sigmoid activation functions are uniformly bounded on and infinitely differentiable.
(2) The boundedness of the weight sequence is valid during training procedure and is the domain of all weights with certain boundary.
(3) The learning rate has an upper bound.
(4) Let denote the weights matrix that consists of all weights and be the order stationary point set of the error function. One necessary condition is that is a finite set.
Then, the influence of regularization on the convergence is derived by using the fractionalorder variational method.
According to (20), is defined as a fractionalorder multivariable function. The proposed fractionalorder BP algorithm is to minimize . Let denote the fractionalorder extreme point of and denotes an admissible point. In addition, is composed of where denotes the weights matrix between the th and th layer when reaches the extreme value. is composed of where corresponds to . The initial weights are random values, so the initial points of weights can be represented as , where is a vector that consists of small parameters , and corresponds to and . If , it means , then , and reaches the extreme value. Thus, the process of training the BP neural networks from a random initial weight to can be treated as the process of training with a random initial value to .
The fractionalorder derivative of on is given aswhere is the fractional order, which is a positive real number.
From (23), we can see that when , if the order differential of with respect to is existent, has a order extreme point and we have
In this case, the output of each layer in the neural networks is still given by (10) and (11) and the input of each layer is turned into the following:
When , we have
Without loss of generality, according to (18), for the th layer of the networks, the order differential of with respect to can be calculated aswhere denotes the column vector .
Since the value of is stochastic, according to variation principle [43], to allow (24) to be set up, a necessary condition is that for every layer of the networks
Secondly, without loss of generality, for we have
To allow (29) to be set up, a necessary condition is
With (28) and (30), the EulerLagrange equation of can be written as
Equation (31) is the necessary condition for the convergence of the proposed fractionalorder BP neural networks with regularization. From (31), we can see that if , then . is the firstorder derivative of in terms of and can be calculated by and input sample . It means that the extreme point of the proposed algorithm is not equal to the extreme point of integerorder BP algorithm or fractionalorder BP algorithm. changes with the different value of and . In addition, it is also clear that the regularization parameter is bounded since the values of input samples and weights are bounded and is a constant during the training process.
5. Experiments
In this section, the following simulations were carried out to evaluate the performance of the presented algorithm. The simulations have been performed on the MNIST handwritten digital dataset. Each digit in the dataset is a 28 × 28 image. Each image is associated with a label from 0 to 9. We divided each image into four parts, which were topleft, bottomleft, bottomright, and topright, and each part was a 14 × 14 matrix. We vectorized each part of the image as a 196 × 1 vector and each label as a 10 × 1 vector.
In order to identify the handwritten digits in MNIST dataset, a neural network with 8 layers was proposed. Figure 1 shows the topological structure of the neural networks. For the first four layers of the network, each layer has 196 external neurons and 32 internal neurons. The outputs of the external neurons are in turn four parts of an image and the outputs of the internal neurons of the first layer are 1. The last four layers have no external neurons. The fifth layer, sixth layer, and seventh layer have 64 internal nodes and the output layer has ten nodes. The activation functions of all neurons except the first layer are sigmoid functions, which can be given as follows:
The MNIST dataset has a total number of 60000 training samples and 10000 testing samples. The simulations demonstrate the performance of the proposed fractionalorder BP neural network with regularization, fractionalorder BP neural network, traditional BP neural network, and traditional BP neural network with regularization. To evaluate the robustness of our proposed network for a small set of training samples, we set the number of training samples to be (10000, 20000, 30000, 40000, 50000, and 60000). Different fractional order derivatives were employed to compute the gradient of error function, where , , , , , , , , , , , , , , , , , , and separately ( corresponds to standard integerorder derivative for the common BP; because if the change of weights after each iteration is 0, and the weights of the neural networks cannot be updated). The learning rate was set to be 3 and the batch size was set to be 100. The number of epochs was 300. Two main metrics—training accuracy and testing accuracy—were used to measure the performance of the results from different networks. Each network was trained 5 times and the average values were calculated.
In order to explore the relationship between the fractional orders and the neural network performance, the fractionalorder neural networks with different orders were trained. Figure 2 shows the results of different networks with different sizes of training set. We can find that when the fractional order exceeds 1.6, both the training and testing accuracies declined rapidly, and when the fractional order , the performances of the fractional BP neural networks were much poorer than that with . The results of and were shown in Table 1 as examples. This result is consistent with that for describing physical problems, and usually the limitation is adopted in the fractionalorder models.

From Figure 2, it can be observed that, with the increase of the size of training set, the performances of the networks were improved visibly. Furthermore, it is also obvious that the training and testing accuracies raised gradually with increasing fractional orders and then reached the peak while equaled or order. After that, the training and testing accuracies began to decline rapidly.
Table 2 shows the optimal orders under training set and testing set separately with different size of training set and it can be noticed that the optimal orders almost concentrated in and . The only exception is that when the number of training samples was 50000, the training accuracy of order 1 was slightly higher than that in or order case. Generally, for the MNIST dataset the performances of fractionalorder BP neural networks are better than integer order.

It also can be seen that, in each case, the training accuracy is much bigger than testing accuracy, which means that the BP neural networks have obvious overfitting phenomenon. To avoid overfitting, the integerorder and fractionalorder BP neural networks with regularization were trained. With different sizes of training set we chose the regularization parameter to be (, , , , , and ). For the fractionalorder neural networks, we chose the fractional order that had highest testing accuracy in previous simulations. When the numbers of training samples were (10000, 20000, 30000, 40000, 50000, and 60000), we separately set the fractional order to be ().
The performance of the proposed fractionalorder BP neural networks with regularization and the performance comparison with integerorder BP neural networks (IOBP), integerorder BP neural networks with regularization, and fractionalorder BP neural networks (FOBP) in terms of training and testing accuracy are shown in Table 3 and the change of the testing accuracy with the iterations was given in Figure 3
 
We use the following formula to calculate improvement: improvement of A compared with B = (AB)÷B. 
In Table 3 and Figure 3, it can be seen that, after the addition of regularization to BP neural networks, the training accuracy is slightly decreased but the testing accuracy significantly increased, which indicated that adding regularization can effectively suppress overfitting and improve the generalization of BP neural networks. Furthermore, it can be noticed that after adding regularization the performance of fractionalorder BP neural network is better than integer order. One important merit of the regularization is that it gained more benefit while the training set is small. The most possible reason is that the network trained with the smallest number of training samples was affected most by the overfitting. With the increase of the training samples, the model gradually changed from overfitting to underfitting, so the improvement of the regularization method became faint.
Then, the stability and convergence of the proposed fractionalorder BP neural networks with regularization are demonstrated in Figures 4 and 5. We used the network with optimal order, which means that the size of training set was 60000, fractionalorder was 11/9, and the regularization parameter was . Figure 4 shows the change of the total error during the training process. Without loss of generality, the change of was randomly selected and Figure 5 shows the change of it during the training process. It is clear to see that and converged fast and stably and were finally close to zero. These observations effectively verify the proposed algorithm is deterministically convergent.
6. Conclusion
In this paper, we applied fractional calculus and regularization method to deep BP neural networks. Different from previous studies, the proposed model had no limitations on the number of layers and the fractionalorder was extended to arbitrary real number bigger than 0. regularization was also imposed into the errorless function. Meanwhile, we analyzed the profits introduced by the regularization on the convergence of this proposed fractionalorder BP network. The numerical results support that the fractionalorder BP neural networks with regularization are deterministically convergent and can effectively avoid the overfitting phenomenon. Then, how to apply fractional calculus to other more complex artificial neural networks is an attracted topic in our future work.
Data Availability
The code of this work can be downloaded at https://github.com/BaoChunhui/DeepfractionalBPneuralnetworks.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
This work was supported in part by the National Key R&D Program of China under Grant 2017YFB0802300, the National Natural Science Foundation of China under Grant 61671312, the Science and Technology Project of Sichuan Province of China under Grant 2018HH0070, and the Strategic Cooperation Project of Sichuan University and Luzhou City under Grant 2015CDLZG22.
References
 M. Kubat, “Neural networks: a comprehensive foundation by Simon Haykin, Macmillan, 1994, ISBN 0023527817.,” The Knowledge Engineering Review, vol. 13, no. 4, pp. 409–412. View at: Publisher Site  Google Scholar
 S. A. Kalogirou, “Applications of artificial neural networks in energy systems: a review,” Energy Conversion and Management, vol. 40, no. 10, pp. 1073–1087, 1999. View at: Publisher Site  Google Scholar
 H. B. Demuth, M. H. Beale, O. De Jess, and M. T. Hagan, Neural network design: Martin Hagan, 2014.
 D. E. Rumelhart, J. L. McClelland, and P. R. Group, Parallel Distributed Processing, vol. 1, MIT press, Cambridge, MA, USA, 1987.
 J. Jia and H. Duan, “Automatic target recognition system for unmanned aerial vehicle via backpropagation artificial neural network,” Aircraft Engineering and Aerospace Technology, vol. 89, no. 1, pp. 145–154, 2017. View at: Publisher Site  Google Scholar
 Z. Wu and H. Wang, Superresolution Reconstruction of SAR Image based on NonLocal Means Denoising Combined with BP Neural Network, 2016, arXiv preprint arXiv:1612.04755.
 E. R. Love, “Fractional derivatives of imaginary order,” Journal Of The London Mathematical SocietySecond Series, vol. 3, pp. 241–259, 1971. View at: Publisher Site  Google Scholar  MathSciNet
 Y. Povstenko, Linear Fractional DiffusionWave Equation for Scientists and Engineers, Birkhäuser, New York, 2015.
 A. McBride and G. Roach, Fractional Calculus (Pitman Research Notes in Mathematics, No 138), Longman Science & Technology, 1986.
 K. Nishimoto, Fractional Calculus: Integrations and Differentiations of Arbitrary Order, University of New Haven Press, New Haven, Conn, USA, 1989. View at: MathSciNet
 I. Podlubny, Fractional differential equations: an introduction to fractional derivatives, fractional differential equations, to methods of their solution and some of their applications, vol. 198, Academic press, 1998.
 N. Özdemir and D. Karadeniz, “Fractional diffusionwave problem in cylindrical coordinates,” Physics Letters A, vol. 372, no. 38, pp. 5968–5972, 2008. View at: Publisher Site  Google Scholar  MathSciNet
 N. Özdemir, O. P. Agrawal, D. Karadeniz, and B. B. İskender, “Analysis of an axissymmetric fractional diffusionwave problem,” Journal of Physics A: Mathematical and Theoretical, vol. 42, no. 35, p. 355208, 2009. View at: Publisher Site  Google Scholar
 Y. Povstenko, “Solutions to the fractional diffusionwave equation in a wedge,” Fractional Calculus and Applied Analysis An International Journal for Theory and Applications, vol. 17, no. 1, pp. 122–135, 2014. View at: Publisher Site  Google Scholar  MathSciNet
 R. L. Bagley and P. J. Torvik, “A theoretical basis for the application of fractional calilus to visoelasticity,” Journal of Rheology, vol. 27, no. 3, pp. 201–210, 1983. View at: Publisher Site  Google Scholar
 D. Baleanu, J. A. T. Machado, and A. C. J. Luo, Fractional Dynamics and Control, Springer, New York, NY, USA, 2012. View at: Publisher Site  MathSciNet
 C. Li and G. Chen, “Chaos in the fractional order Chen system and its control,” Chaos, Solitons $ Fractals, vol. 22, pp. 549–554, 2004. View at: Google Scholar
 C. A. Monje, B. M. Vinagre, V. Feliu, and Y. Chen, “Tuning and autotuning of fractional order controllers for industry applications,” Control Engineering Practice, vol. 16, no. 7, pp. 798–812, 2008. View at: Publisher Site  Google Scholar
 L. B. Almeida, “Fractional fourier transform and timefrequency representations,” IEEE Transactions on Signal Processing, vol. 42, no. 11, pp. 3084–3091, 1994. View at: Publisher Site  Google Scholar
 M. S. Aslam and M. A. Z. Raja, “A new adaptive strategy to improve online secondary path modeling in active noise control systems using fractional signal processing approach,” Signal Processing, vol. 107, pp. 433–443, 2015. View at: Publisher Site  Google Scholar
 R. Panda and M. Dash, “Fractional generalized splines and signal processing,” Signal Processing, vol. 86, no. 9, pp. 2340–2350, 2006. View at: Publisher Site  Google Scholar
 M. Xu, J. Yang, D. Zhao, and H. Zhao, “An imageenhancement method based on variableorder fractional differential operators,” BioMedical Materials and Engineering, vol. 26, pp. S1325–S1333, 2015. View at: Publisher Site  Google Scholar
 Y.F. Pu, N. Zhang, Y. Zhang, and J.L. Zhou, “A texture image denoising approach based on fractional developmental mathematics,” PAA. Pattern Analysis and Applications, vol. 19, no. 2, pp. 427–445, 2016. View at: Publisher Site  Google Scholar  MathSciNet
 Y.F. Pu, J.L. Zhou, and X. Yuan, “Fractional differential mask: a fractional differentialbased approach for multiscale texture enhancement,” IEEE Transactions on Image Processing, vol. 19, no. 2, pp. 491–511, 2010. View at: Publisher Site  Google Scholar  MathSciNet
 J. Bai and X.C. Feng, “Fractionalorder anisotropic diffusion for image denoising,” IEEE Transactions on Image Processing, vol. 16, no. 10, pp. 2492–2502, 2007. View at: Publisher Site  Google Scholar  MathSciNet
 Y. Zhang, W. Zhang, Y. Lei, and J. Zhou, “Fewview image reconstruction with fractionalorder total variation,” Journal of the Optical Society of America A: Optics, Image Science & Vision, vol. 31, no. 5, pp. 981–995, 2014. View at: Publisher Site  Google Scholar
 Y. Zhang, Y. Wang, W. Zhang, F. Lin, Y. Pu, and J. Zhou, “Statistical iterative reconstruction using adaptive fractional order regularization,” Biomedical Optics Express, vol. 7, no. 3, pp. 1015–1029, 2016. View at: Publisher Site  Google Scholar
 Y. Zhang, Y.F. Pu, J.R. Hu, Y. Liu, Q.L. Chen, and J.L. Zhou, “Efficient CT metal artifact reduction based on fractionalorder curvature diffusion,” Computational and Mathematical Methods in Medicine, Art. ID 173748, 9 pages, 2011. View at: Google Scholar  MathSciNet
 Y.F. Pu, Z. Yi, and J.L. Zhou, “Defense Against Chip Cloning Attacks Based on Fractional Hopfield Neural Networks,” International Journal of Neural Systems, vol. 27, no. 4, Article ID 1750003, 2017. View at: Publisher Site  Google Scholar
 E. Kaslik and S. Sivasundaram, “Dynamics of fractionalorder neural networks,” in Proceedings of the 2011 International Joint Conference on Neural Network, IJCNN 2011, pp. 611–618, usa, August 2011. View at: Google Scholar
 Y.F. Pu, Z. Yi, and J.L. Zhou, “Fractional Hopfield neural networks: fractional dynamic associative recurrent neural networks,” IEEE Transactions on Neural Networks and Learning Systems, vol. 28, no. 10, pp. 2319–2333, 2017. View at: Publisher Site  Google Scholar  MathSciNet
 C. Song and J. Cao, “Dynamics in fractionalorder neural networks,” Neurocomputing, vol. 142, pp. 494–498, 2014. View at: Publisher Site  Google Scholar
 J. Wang, Y. Wen, Y. Gou, Z. Ye, and H. Chen, “Fractionalorder gradient descent learning of BP neural networks with Caputo derivative,” Neural Networks, vol. 89, pp. 19–30, 2017. View at: Publisher Site  Google Scholar
 R. Rakkiyappan, R. Sivaranjani, G. Velmurugan, and J. Cao, “Analysis of global O (t a) stability and global asymptotical periodicity for a class of fractionalorder complexvalued neural networks with time varying delays,” Neural Networks, vol. 77, pp. 51–69, 2016. View at: Google Scholar
 H. Wang, Y. Yu, and G. Wen, “Stability analysis of fractionalorder Hopfield neural networks with time delays,” Neural Networks, vol. 55, pp. 98–109, 2014. View at: Publisher Site  Google Scholar
 H. Wang, Y. Yu, G. Wen, S. Zhang, and J. Yu, “Global stability analysis of fractionalorder Hopfield neural networks with time delay,” Neurocomputing, vol. 154, pp. 15–23, 2015. View at: Publisher Site  Google Scholar
 A. Boroomand and M. B. Menhaj, “FractionalOrder Hopfield Neural Networks,” in Advances in NeuroInformation Processing, vol. 5506 of Lecture Notes in Computer Science, pp. 883–890, Springer Berlin Heidelberg, Berlin, Heidelberg, 2009. View at: Publisher Site  Google Scholar
 Y.F. Pu, J.L. Zhou, Y. Zhang, N. Zhang, G. Huang, and P. Siarry, “Fractional extreme value adaptive training method: fractional steepest descent approach,” IEEE Transactions on Neural Networks and Learning Systems, vol. 26, no. 4, pp. 653–662, 2015. View at: Publisher Site  Google Scholar  MathSciNet
 H. Shao and G. Zheng, “Boundedness and convergence of online gradient method with penalty and momentum,” Neurocomputing, vol. 74, no. 5, pp. 765–770, 2011. View at: Publisher Site  Google Scholar
 W. Wu, G. Feng, Z. Li, and Y. Xu, “Deterministic convergence of an online gradient method for BP neural networks,” IEEE Transactions on Neural Networks and Learning Systems, vol. 16, no. 3, pp. 533–540, 2005. View at: Publisher Site  Google Scholar
 W. Wu, J. Wang, M. Cheng, and Z. Li, “Convergence analysis of online gradient method for BP neural networks,” Neural Networks, vol. 24, no. 1, pp. 91–98, 2011. View at: Publisher Site  Google Scholar
 H. Zhang, W. Wu, F. Liu, and M. Yao, “Boundedness and convergence of online gadient method with penalty for feedforward neural networks,” IEEE Transactions on Neural Networks and Learning Systems, vol. 20, no. 6, pp. 1050–1054, 2009. View at: Publisher Site  Google Scholar
 G. Leitmann, The calculus of variations and optimal control: an introduction, vol. 24, Springer Science & Business Media, 2013.
Copyright
Copyright © 2018 Chunhui Bao et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.