Approximate greatest descent in neural network optimization

King Hann Lim; Hong Hui Tan; Hendra G. Harno

doi:10.3934/naco.2018021

Article Contents

2018, Volume 8, Issue 3: 327-336. Doi: 10.3934/naco.2018021

This issue Previous Article On a two-phase approximate greatest descent method for nonlinear optimization with equality constraints Next Article Homotopy perturbation method and Chebyshev polynomials for solving a class of singular and hypersingular integral equations

Approximate greatest descent in neural network optimization

1.
Faculty of Engineering and Science, Curtin University Malaysia, Malaysia
2.
Department of Aerospace and Software Engineering, Gyeongsang National University, South Korea

* Corresponding author: King Hann Lim
* Corresponding author: King Hann Lim

Received: April 2017

Revised: December 2017

Published: June 2018

Abstract / Introduction Full Text(HTML) Figure(3) / Table(1) Related Papers Cited by

Abstract

Numerical optimization is required in artificial neural network to update weights iteratively for learning capability. In this paper, we propose the use of Approximate Greatest Descent (AGD) algorithm to optimize neural network weights using long-term backpropagation manner. The modification and development of AGD into stochastic diagonal AGD (SDAGD) algorithm could improve the learning ability and structural simplicity for deep learning neural networks. It is derived from the operation of a multi-stage decision control system which consists of two phases: (1) when local search region does not contain the minimum point, iteration shall be defined at the boundary of the local search region, (2) when local region contains the minimum point, Newton method is approximated for faster convergence. The integration of SDAGD into Multilayered perceptron (MLP) network is investigated with the goal of improving the learning ability and structural simplicity. Simulation results showed that two-layer MLP with SDAGD achieved a misclassification rate of 9.4% on a smaller mixed national institute of national and technology (MNIST) dataset. MNIST is a database equipped with handwritten digits images suitable for algorithm prototyping in artificial neural networks.

Keywords:

Mathematics Subject Classification: Primary: 58F15, 58F17; Secondary: 53C35.

Citation:

FullText(HTML)

Figure 1. Structure of two layer multilayer perceptron.

Download: Full-size image PowerPoint slide

Figure 2. AGD iteration from initial point to minimum point.

Download: Full-size image PowerPoint slide

Figure 3. AGD iteration from initial point to minimum point.

Download: Full-size image PowerPoint slide

Table 1. Comparison of MCR and MSE between three optimization techniques in neural network.

Training Algorithm Training MCR (%) Testing MCR (%) MSE

SGD 8.22 12.14 0.40

SDLM 8.86 10.19 0.32

SDAGD 6.46 9.40 0.21

| Show Table

DownLoad: CSV

Related Papers

Cited by

References

	S. Amari , H. Park and K. Fukumizu , Adaptive method of realizing natural gradient learning for multilayer perceptron, Neural Compt., 12 (2000) , 436-444. doi: 10.1162/089976600300015420.
	S. Becker and Y. LeCun , Improving the convergence of backpropagation learning with second order methods, Proc. of the Con. Models Summer School, (1988) , 29-37.
	Y. Bengio , Learning deep architectures for AI, Foundations and trends in Machine Learning, 2 (2009) , 1-127.
	L. Bottou , Large-scale machine learning with stochastic gradient descent, Proc. of COMPSTAT, (2010) , 177-186.
	X. Glorot and Y. Bengio , Understanding the difficulty of training deep feedforward neural networks, Aistats, 9 (2010) , 249-256.
	B. S. Goh , Greatest descent algorithms in unconstrained optimization, J. Optim. Theory Appl., 142 (2009) , 275-289. doi: 10.1007/s10957-009-9533-4.
	B. S. Goh , Numerical method in optimization as a multi-stage decision control system, Latest Advances in Systems Science and Computational Intelligence, (2012) , 25-30.
	Y. LeCun, L. Bottou, G. B. Orr and K. R. Müller, Efficient backprop, Neural Networks: Tricks of the Trade, Springer, (2012), 9–48.
	Y. LeCun , Y. Bengio and G. Hinton , Deep learning, Nature, 521 (2015) , 436-444. doi: 10.1038/nature14539.
	Y. LeCun , L. Bottou , Y. Bengio and P. Haffner , Gradient-based learning applied to document recognition, Proc. IEEE, 86 (1998) , 2278-2323. doi: 10.1109/5.726791.
	K. H. Lim , K. P. Seng , L. M. Ang and S. W. Chin , Lyapunov theory-based multilayered neural network, IEEE Transactions on Circuits and Systems II: Express Briefs, 4 (2009) , 305-309.
	J. Nocedal and S. Wright, Numerical Optimization, 2nd ed., Springer, 2006.
	J. R. Shewchuk, An Introduction to the Conjugate Gradient Method Without the Agonizing Pain, Tech. Rep. C. Carnegie-Mellon Univ., 1994.
	J. Sohl-Dickstein , B. Poole and S. Ganguli , Fast large-scale optimization by unifying stochastic gradient and quasi-Newton methods, Procs. 31st Int. Conf. Mach. Learn, (2014) , 604-612.
	D. Stutz, Introduction to Neural Networks, Selected Topics in Human Language Technology and Pattern Recognition WS 12/14, 2014.
	H. H. Tan , K. H. Lim and H. G. Harno , Stochastic diagonal approximate greatest descent in neural networks, 2017 International Joint Conference on Neural Networks (IJCNN), (2007) , 1895-1898. doi: 10.1109/IJCNN.2017.7966081.