In a neural network based character recognition system it is important to choose a training algorithm with high generalization ability. In this paper, we apply three different multilayer feedforward training algorithm...
详细信息
ISBN:
(纸本)0780371011
In a neural network based character recognition system it is important to choose a training algorithm with high generalization ability. In this paper, we apply three different multilayer feedforward training algorithms namely, Backpropagation, double Backpropagation and weight smoothing algorithm in a neural network based invariant character recognition model proposed in [16]. The model consists of a preprocessor and a classifier. The preprocessor extracts geometrical features of the input character and passes the feature values through a Rapid Transform block which performs cyclic shift invariant transform on its inputs. The classifier is a neural network classifier. Simulation results with 26 English capital letters show that the recognition system achieves best performance with significantly high recognition rate when trained with weight smoothing learning algorithm.
Although the backpropagation (BP) scheme is widely used as a learning algorithm for multilayered neural networks, the learning speed of the BP algorithm to obtain acceptable errors is unsatisfactory in spite of sonic ...
详细信息
Although the backpropagation (BP) scheme is widely used as a learning algorithm for multilayered neural networks, the learning speed of the BP algorithm to obtain acceptable errors is unsatisfactory in spite of sonic improvements such as introduction of a momentum factor and an adaptive learning rate in the weight adjustment. To solve this problem, a fast learning algorithm based on the extended Kalman filter (EKF) is presented and fortunately its computational complexity has been reduced by some simplifications. In general, however, the Kalman filtering algorithm is well known to be sensitive to the nature of noises which is generally assumed to be Gaussian. In addition, the H-infinity theory suggests that the maximum energy gain of the Kalman algorithm from disturbances (initial state, system, and observation noises) to the estimation error has no upper bound. That is, the Kalman filtering algorithm has a poor robustness to such disturbances. Therefore, the EKF-based learning algorithms should be further improved to enhance the robustness to variations in the initial values of link weights and thresholds as well as to the nature of noises. The aim of this paper is to propose H-infinity-learning as a novel learning rule and to derive new globally and locally optimized learning algorithms based on H.-learning. Their learning behavior is analyzed from various points of view using computer simulations. The derived algorithms are also compared, in performance and computational cost, with the conventional BP and EKF learning algorithms.
An adaptive wavelet-based neural network is proposed by combining wavelet transformation with neural network theory in this paper. The parameters of scale and dulation of the wavelet are adjusted adaptively according ...
详细信息
An adaptive wavelet-based neural network is proposed by combining wavelet transformation with neural network theory in this paper. The parameters of scale and dulation of the wavelet are adjusted adaptively according to signal's characteristic in the learning process, so that the feature of the signal could be extracted to a large extent. Classification of mechanical faults based on adaptive wavelet-based neural network is also researched. The result of an example of bearing fault classification demonstrates that the neural network can classify fault accurately and reliably.
We study the uniform graph partitioning problem using the learning algorithm proposed by one of us. We discuss the characteristics of the learning algorithm and compare the performance of the algorithm empirically wit...
详细信息
We study the uniform graph partitioning problem using the learning algorithm proposed by one of us. We discuss the characteristics of the learning algorithm and compare the performance of the algorithm empirically with the Kernighan-Lin algorithm on a range of instances. Even with a simple implementation, the learning algorithm is capable of producing very good results.
The advantage of TSK fussy system (Sugeno-Tanaka fuzzy system) is combined into CMAC (Cerebellar Model Articulation Controller) neural network, and therefore the improved CMAC neural network is presented and its learn...
详细信息
The advantage of TSK fussy system (Sugeno-Tanaka fuzzy system) is combined into CMAC (Cerebellar Model Articulation Controller) neural network, and therefore the improved CMAC neural network is presented and its learning algorithm. Our experiment results have showed that the Improved-CMAC outperforms CMAC in learning precision.
The advantage of TSK Fuzzy System is combined into FCP (Fuzzy Counter-propagation) and the improved FCP neural network (Improved-FCP) and its learning algorithm are further presented. In the end of this paper, the exp...
详细信息
The advantage of TSK Fuzzy System is combined into FCP (Fuzzy Counter-propagation) and the improved FCP neural network (Improved-FCP) and its learning algorithm are further presented. In the end of this paper, the experiment results show that the Improved-FCP has much better approximation accuracy than FCP.
In this paper the models discussed by Cohen are extended by introducing an input term. This allows the resulting models to be utilized for system identification tasks. This approach gives a direct way to encode qualit...
详细信息
In this paper the models discussed by Cohen are extended by introducing an input term. This allows the resulting models to be utilized for system identification tasks. This approach gives a direct way to encode qualitative information such as attractor dimension into the model. We prove that this model is stable in the sense that a bounded input leads to a bounded state when a minor restriction is imposed on the Lyapunov function. By employing this stability result, we are able to find a learning algorithm which guarantees convergence to a set of parameters for which the error between the model trajectories and the desired trajectories vanishes. (C) 1998 Elsevier Science Ltd. All lights reserved.
The paper presents a robust fault diagnosis scheme for detecting and approximating state and output faults occurring in a class of nonlinear multiinput-multioutput dynamical systems, Changes in the system dynamics due...
详细信息
The paper presents a robust fault diagnosis scheme for detecting and approximating state and output faults occurring in a class of nonlinear multiinput-multioutput dynamical systems, Changes in the system dynamics due to a fault are modeled as nonlinear functions of the control input and measured output variables, Both state and output faults can be modeled as slowly developing (incipient) or abrupt, with each component of the state/output fault vector being represented by a separate time profile. The robust fault diagnosis scheme utilizes on-line approximators and adaptive nonlinear filtering techniques to obtain estimates of the fault functions. Robustness with respect to modeling uncertainties, fault sensitivity and stability properties of the learning scheme are rigorously derived and the theoretical results are illustrated by a simulation example of a fourth-order satellite model.
An algorithm is proposed for training the single-layered perceptron. The algorithm follows successive steepest descent directions with respect to the perceptron cost function, taking care not to increase the number of...
详细信息
An algorithm is proposed for training the single-layered perceptron. The algorithm follows successive steepest descent directions with respect to the perceptron cost function, taking care not to increase the number of misclassified patterns. The problem of finding these directions is stated as a quadratic programming task, to which a fast and effective solution is proposed. The resulting algorithm has no free parameters and therefore no heuristics are involved in its application. It is proved that the algorithm always converges in a finite number of steps. For linearly separable problems, it always finds a hyperplane that completely separates patterns belonging to different categories. Termination of the algorithm without separating all given patterns means that the presented set of patterns is indeed linearly inseparable. Thus the algorithm provides a natural criterion for linear separability. Compared to other state of the art algorithms, the proposed method exhibits substantially improved speed, as demonstrated in a number of demanding benchmark classification tasks. (C) 2000 Published by Elsevier Science Ltd. All rights reserved.
The learning approach to fault diagnosis provides a methodology for designing monitoring architectures which can he used for detection, identification and accommodation of failures in dynamical systems. This paper con...
详细信息
The learning approach to fault diagnosis provides a methodology for designing monitoring architectures which can he used for detection, identification and accommodation of failures in dynamical systems. This paper considers the issues of detectability conditions and detection time in a nonlinear fault diagnosis scheme based on the learning approach. First, conditions are derived to characterize the range of detectable faults. Then, nonconservative upper bounds are computed for the detection time of incipient and abrupt faults. It is shown that the detection time bound decreases monotonically as the values of certain design parameters increase. The theoretical results are illustrated by a simulation example of a second-order system.
暂无评论