HIDDEN CONTROL NEURAL ARCHITECTURE MODELING OF NONLINEAR TIME-VARYING SYSTEMS AND ITS APPLICATIONS

Multilayered neural networks have been proposed recently for nonlinear prediction and system modeling [1]-[6]. Although neural networks have proven.successful for modeling time invariant nonlinear systems, it has been difficult to apply them to complicated nonstationary signals, such as speech because they are unable to characterize temporal variability. In this paper we address this problem by proposing a network architecture, called ''hidden control neural network'' (HCNN) [4], for modeling signals generated by nonlinear dynamical systems with restricted time variability. Our approach is to allow the mapping that is implemented by a multilayered neural network to change with time as a function of an additional control input signal. This network is trained using an algorithm that is based on ''back-propagation'' and segmentation algorithms for estimating the unknown control together with the network's parameters. This network wa$ applied to three different tasks: segmentation and modeling of a signal produced by a time-varying nonlinear system; speaker-independent recognition of spoken connected digit; and on-line recognition of handwritten characters. The results of these experiments demonstrate the ability of HCNN to learn time-varying nonlinear dynamics, and its potential for high performance recognition of signals, produced by time-varying sources, such as spoken utterances, and written characters.