The Cerebellar Model Arithmetic Computer (CMAC) proposed by Albus is a neural network that imitates the human cerebellum, and it is basically a table lookup technique for representing nonlinear functions. The CMAC can...
详细信息
The Cerebellar Model Arithmetic Computer (CMAC) proposed by Albus is a neural network that imitates the human cerebellum, and it is basically a table lookup technique for representing nonlinear functions. The CMAC can approximate a wide variety of nonlinear functions by learning. However, the CMAC requires a number of learning iterations to achieve the adequate approximation accuracy, because CMAC learning algorithm is based on the Least Mean Square method. In this paper, a CMAC learning algorithm based on the Kalman filter is proposed. Using this algorithm, the number of learning iterations is reduced, and the comparable approximation accuracy is achieved as compared to the system using conventional learning algorithm. Generally, the learning algorithm based on the Kalman filter requires much larger computational quantity than that required by algorithms based on the Least Mean Square method. For CMAC system equations contain a sparse matrix, the computational quantity of the proposed learning algorithm can be reduced. Furthermore, since two CMAC weights being in the far place from each other can be considered to have little correlation, the number of weights that should be updated for an learning iteration can be reduced. This reduces the computational quantity of the learning algorithm. Computer simulation results for the modeling problems are presented.
We are now developing a brain computer with algorithm acquisition function, where a two-level structure is introduced to connect pattern with (meta-)symbol, because we know how to realize algorithm acquisition on symb...
详细信息
ISBN:
(纸本)0819431184
We are now developing a brain computer with algorithm acquisition function, where a two-level structure is introduced to connect pattern with (meta-)symbol, because we know how to realize algorithm acquisition on symbols. At Level 1 we use a conventional learning method on neural networks, but, at level 2;we develop a new learning algorithm AST (Abstract State Transition algorithm), where an automaton-like algorithm with a neural, network learning is introduced. This is enough powerful to realize an automatic algorithm acquisition. We will state a two-level structure and the AST learning algorithm. We focus on real-time image understanding which is a realization of human brain with eyes. We will summarize the features of our developing artificial brain system as follows;I) System, for Meta-Symbol as well as Pattern. 2) Architecture with algorithm Acquisition Function. 3) Cognitive Memory Model as well as Biological Memory Model. To realize an artificial memory model to satisfy the features of 1)-3), we introduce a two-level architecture, where the Meta-Symbol is introduced at Level 2 while the Pattern is used for Level 1 as usual.
The relationship between generalization and learning error function is studied, while the training data is contaminated by noise. The random neural network is considered and the K-L information distance is used to est...
详细信息
The relationship between generalization and learning error function is studied, while the training data is contaminated by noise. The random neural network is considered and the K-L information distance is used to estimate the error of the neural network. It is proved that the K-L information distance keeps the consistence with the generalization. The over-fitting is analyzed based on the K-L distance. A new learning error function is proposed, and the generalization error is estimated if this learning error function is applied. A simulation example is given to show that the proposed learning error function exactly improves the generalization.
The GMDH network is a learning machine based on the principle of heuristic self-organization. In this paper, use of the GMDH network for predicting the testing progress of software products is discussed. The fundament...
详细信息
The GMDH network is a learning machine based on the principle of heuristic self-organization. In this paper, use of the GMDH network for predicting the testing progress of software products is discussed. The fundamental GMDH and the improved GMDH using the AIC as the evaluation criterion are introduced for estimating the fault-occurrence times observed in the testing of software. Finally, in a numerical example, the GMDH network, an existing software reliability growth model, and a multilayered neural network are compared from the viewpoint of predicting performance. As a result, it is shown that the GMDH network overcomes the problem of determining an adequate network size in using a multilayered neural network and, in addition, provides a more accurate measure in evaluating software reliability than other prediction models. (C) 1999 Scripta Technica.
This paper summarizes a series of our researches on the generalized Hopfield network for associative ***,the stability of the generalized Hopfield network is *** the memory capacity of the generalized Hopfield network...
详细信息
This paper summarizes a series of our researches on the generalized Hopfield network for associative ***,the stability of the generalized Hopfield network is *** the memory capacity of the generalized Hopfield network is discussed and a lower and an upper bounds of it are ***,a learning algorithm referred to as the object perceptron learning algorithm is proposed on the generalized Hopfield network for associative memory.
Stochastic learning automata are used to 'grow' quality of service bounded multicast trees in a dynamic membership environment. It is found that learning automata, which use minimal state information and requi...
详细信息
Stochastic learning automata are used to 'grow' quality of service bounded multicast trees in a dynamic membership environment. It is found that learning automata, which use minimal state information and require only local connectivity knowledge, provide superior results to alternative shortest path approaches by learning to minimise the number of blocked connection attempts via load balancing.
This paper describes a new type of learning control method for precision velocity control of servomotors suffering from significant disturbance torque. The disturbance torque under consideration is assumed to be perio...
详细信息
This paper describes a new type of learning control method for precision velocity control of servomotors suffering from significant disturbance torque. The disturbance torque under consideration is assumed to be periodic in time and nonlinear in system states, but possibly non-Lipschitzian. Based on the property that the learning system tends to oscillate in the steady state, the proposed learning algorithm iteratively generates a feedforward input to cancel the effect of the disturbance torque. Thereby, it can eventually drive the steady-state velocity error to zero. In order to demonstrate the generality of the proposed method, we present a rigorous analysis for the convergence of the proposed learning algorithm. The effectiveness of the proposed method is demonstrated by simulation and experiment.
Product life cycle theory has been a key organizing principle in studies of technical innovation over the last 20 years and is promoted by leading management theorists as a tool for strategic decision making. This pap...
详细信息
Product life cycle theory has been a key organizing principle in studies of technical innovation over the last 20 years and is promoted by leading management theorists as a tool for strategic decision making. This paper critically appraises the dominant design concept that lies at the heart of this theory. In so doing it seriously questions a basic premise adopted by numerous economists;that the dynamics of technological change can be understood through the examination of the technological artefact. We put forward an alternative explanation for the patterns of product innovation observed in infant industries. This alternative views technological innovation as a coupled, second-order learning system comprising a population of consumers and a population of firms. Within this approach the artefact is viewed in a quite different light. Rather than being an object in itself with its own internal drives and dynamics, it is a mediating device. The form of the mediating device alters over time due to changes in the external factors acting upon it;that is to say, its form changes as a consequence of the formation and development of user preferences through consumer learning and of the technological experimentation and beaming of firms. Turning specifically to the dominant design concept, we argue that this is but one possible outcome of the innovation system. It is shown that, while the system tends to converge towards a limited number of design configurations, there is no reason to expect it to stabilize around a single design. It may also stabilize through a process of market differentiation leading to the emergence of distinct niches. Observations drawn from the markets for cameras, road vehicles, amplification systems and personal computers support the contention that the dominant design outcome is a 'special case' - a case of convergence to a single market niche. The theoretical and empirical analysis is formalized through a simulation model of the innovation system. Th
During the last decade, researchers have applied neural networks to a multitude of difficult tasks which would normally require human intelligence. In particular, perceptrons are used to classify patterns into differe...
详细信息
ISBN:
(纸本)0780344235
During the last decade, researchers have applied neural networks to a multitude of difficult tasks which would normally require human intelligence. In particular, perceptrons are used to classify patterns into different classes. Recently, several researchers introduced a novel class of artificial neural networks, called morphological neural networks. In this new theory, the first step in computing the next state of a neuron of in performing the next layer neural network computation involves the nonlinear operation of adding neural values and their synaptic strengths followed by forming the maximum of the results. We have shown in previous papers that the properties of morphological neural networks differ drastically from those of traditional neural network models. In this paper, we introduce a learning algorithm for multilayer morphological;perceptrons which is capable of solving arbitrary classification problems of patterns into two classes.
This paper presents an overview of a learning methodology for detecting, identifying and accommodating faults in nonlinear dynamic systems. The main idea behind this approach is to monitor the plant for any off-nomina...
详细信息
ISBN:
(纸本)0780344235
This paper presents an overview of a learning methodology for detecting, identifying and accommodating faults in nonlinear dynamic systems. The main idea behind this approach is to monitor the plant for any off-nominal behavior due to faults utilizing a neural network or other types of online approximators. In the presence of a failure, the neural network can be used as an estimate of the nonlinear fault function for fault identification and accommodation purposes. Furthermore, during the initial stage of monitoring, the learning capabilities of the neural network can be used to learn the modeling errors, thereby enhancing the robustness properties of the fault diagnosis scheme.
暂无评论