Recurrent neural networks training with stable risk-sensitive Kalman filter algorithm

Compared to normal learning algorithms, for example backpropagation, Kalman filter-based algorithm has some better properties, such as faster convergence. In this paper, Kalman filter is modified with a risk-sensitive cost criterion, we call it as risk-sensitive Kalman filter. This new algorithm is applied to train recurrent neural networks for nonlinear system identification. Input-to-state stability is used to prove that the risk-sensitive Kalman filter training is stable. The contributions of this paper are: 1) the risk-sensitive Kalman filter is used for the state-space recurrent neural networks training, 2) the stability of the risk-sensitive Kalman filter is proved.

[1]  C. Leung,et al.  Use of periodic and monotonic activation functions in multilayer feedforward neural networks trained by extended Kalman filter algorithm , 2002 .

[2]  Sharad Singhal,et al.  Training Multilayer Perceptrons with the Extende Kalman Algorithm , 1988, NIPS.

[3]  Kumpati S. Narendra,et al.  Identification and control of dynamical systems using neural networks , 1990, IEEE Trans. Neural Networks.

[4]  Lee A. Feldkamp,et al.  Neurocontrol of nonlinear dynamical systems with Kalman filter trained recurrent networks , 1994, IEEE Trans. Neural Networks.

[5]  Ian R. Petersen,et al.  Robustness and risk-sensitive filtering , 2002, IEEE Trans. Autom. Control..

[6]  Fahmida N. Chowdhury.  A new approach to real‐time training of dynamic neural networks , 2003 .

[7]  Ravi N. Banavar,et al.  Risk-Sensitive Filters for Recursive Estimation of Motion From Images , 1998, IEEE Trans. Pattern Anal. Mach. Intell..

[8]  James T. Lo,et al.  Existence and uniqueness of risk-sensitive estimates , 2002, IEEE Trans. Autom. Control..

[9]  P. Whittle.  Risk-Sensitive Optimal Control , 1990 .

[10]  Mark E. Oxley,et al.  Comparative Analysis of Backpropagation and the Extended Kalman Filter for Training Multilayer Perceptrons , 1992, IEEE Trans. Pattern Anal. Mach. Intell..

[11]  Konrad Reif,et al.  The extended Kalman filter as an exponential observer for nonlinear systems , 1999, IEEE Trans. Signal Process..

[12]  John B. Moore,et al.  Finite-dimensional risk-sensitive filters and smoothers for discrete-time nonlinear systems , 1999, IEEE Trans. Autom. Control..

[13]  Amir F. Atiya,et al.  An algorithmic approach to adaptive state filtering using recurrent neural networks , 2001, IEEE Trans. Neural Networks.

[14]  Francesco Palmieri,et al.  Optimal filtering algorithms for fast learning in feedforward neural networks , 1992, Neural Networks.

[15]  Hideaki Sakai,et al.  A real-time learning algorithm for a multilayered neural network based on the extended Kalman filter , 1992, IEEE Trans. Signal Process..

[16]  Richard D. Braatz,et al.  On the "Identification and control of dynamical systems using neural networks" , 1997, IEEE Trans. Neural Networks.

[17]  Wen Yu,et al.  Nonlinear system identification using discrete-time recurrent neural networks with stable learning algorithms , 2004, Inf. Sci..