Convergence analysis of the   weighted state space search algorithm   for  recurrent neural networks

Leong-Kwan Li; Sally Shao

doi:10.3934/naco.2014.4.193

Article Contents

2014, Volume 4, Issue 3: 193-207. Doi: 10.3934/naco.2014.4.193

This issue Previous Article Robust control design of autonomous bicycle kinematics Next Article Two-step methods for image zooming using duality strategies

Convergence analysis of the weighted state space search algorithm for recurrent neural networks

Leong-Kwan Li^1, and
Sally Shao^2,

1.
Department of Applied Mathematics, The Hong Kong Polytechnic University, Kowloon, Hong Kong
2.
Department of Mathematics, Cleveland State University, Cleveland, OH 44115

Received: March 31, 2013

Revised: June 30, 2014

Published: August 2014

Abstract Related Papers Cited by

Abstract

Recurrent neural networks (RNNs) have emerged as a promising tool in modeling nonlinear dynamical systems. The convergence is one of the most important issues of concern among the dynamical properties for the RNNs in practical applications. The reason is that the viability of many applications of RNNs depends on their convergence properties. We study in this paper the convergence properties of the weighted state space search algorithm (WSSSA) -- a derivative-free and non-random learning algorithm which searches the neighborhood of the target trajectory in the state space instead of the parameter space. Because there is no computation of partial derivatives involved, the WSSSA has a couple of salient features such as simple, fast and cost effective. In this study we provide a necessary and sufficient condition that required for the convergence of the WSSSA. Restrictions are offered that may help assure convergence of the of the WSSSA to the desired solution. The asymptotic rate of convergence is also analyzed. Our study gives insights into the problem and provides useful information for the actual design of the RNNs. A numerical example is given to support the theoretical analysis and to demonstrate that it is applicable to many applications.

Keywords:

Mathematics Subject Classification: Primary: 68T05, 65B99; Secondary: 68Q32.

Citation:

Related Papers

Cited by

References

[1]	A. F. Atiya and A. G. Parlos, New results on recurrent network training: Unifying the algorithms and accelerating convergence, IEEE Transcations on Neural Networks, 11 (2000), 697-709.
[2]	R. A. Conn, K. Scheinberg and N. L. Vicente, Introduction to Derivative-free Optimization, SIAM, Philadelphia, 2009.doi: 10.1137/1.9780898718768.
[3]	L. Jin, N. Nikifork and M. M. Gupta, Absolute stability conditions for discrete-time neural networks, IEEE Tranc. Neural Networks, 5 (1994), 954-964.
[4]	L. K. Li, Learning sunspot series dynamics by recurrent neural networks, Advances in Data Mining and Modeling (eds. W. K. Ching and K. P. Ng), World Science (2003), 107-115.
[5]	L. K. Li and S. Shao, Dynamic properties of recurrent neural networks and its approximations, International Journal of Pure and Applied Mathematics, 39 (2007), 545-562.
[6]	L. K. Li, S. Shao and T. Zheleva, A state space search algorithm and its application to learn the short-term foreign exchange rates, Applied Mathematical Sciences, 2 (2008), 1705-1728.
[7]	L. K. Li, Sally Shao and K. F. Cedric Yiu, Nonlinear dynamical system modeling via recurrent neural networks and a weighted wtate space search algorithm, Journal of Industrial and Management Optimization, 7 (2011), 385-400.doi: 10.3934/jimo.2011.7.385.
[8]	Q. Liu and J. Wang, Finite-time convergent recurrent neural network with a hard-liming activation function for constrained optimization with piecewise-linear objective functions, IEEE Transactions on Neural Networks, 22 (2011), 601-613.
[9]	D. T. Mirikitani and N. Nikolaev, Recursive Bayesian recurrent neural networks for time-series modeling, IEEE Transactions on Neural Networks, 2 (2010), 262-274.
[10]	Q. Song, On the weight convergence of Elman networks, IEEE Transactions on Neural Networks, 21 (2010), 463-480.
[11]	X. Wang and E. K. Blum, Discrete-time versus continuous-time models of neural networks, Journal of Computer and System Sciences, 45 (1992), 1-19.doi: 10.1016/0022-0000(92)90038-K.
[12]	X. Wang and H. Huang, Convergence Study in Extended Kalman Filter-based Training of Recurrent Neural Networks, IEEE Trans. on Neural Networks, 22 (2011), 588-600.
[13]	L. Xu and W. Liu, A new recurrent neural network adaptive approach for host-gate way rate control protocol within intranets using ATM ABR service, Journal of Industrial and Management Optimization, 1 (2005), 389-404.doi: 10.3934/jimo.2005.1.337.
[14]	F. Xu and Z. Yi, Convergence Analysis of a class of simplified background netural networks with subnetworks, Neurocomputing, 74 (2011), 3877-3883.
[15]	J. Yao and C. L. Tan, A case study on using neural networks to perform technical forecasting of forex, Neural Computation, 34 (2000), 79-98.
[16]	K. F. C. Yiu, S. Wang, K. L. Teo and A. C. Tsoi, Nonlinear System modeling via knot-optimizing B-splines networks, IEEE Transactions on Neural Networks, 12 (2001), 1013-1022.
[17]	Y. Zhang and K. K. Tan, Convergence Analysis of Recurrent Neural Networks. Kluwer, Norwell, MA, (2004).doi: 10.1007/978-1-4757-3819-3.
[18]	L. Zhang and Z. Yi., Selectable and unselectable sets of neurons in recurrent neural networks with saturated piecewise linear transfer function, IEEE Transactions on Neural Networks, 22 (2011), 1021-1031.