A neural network approach-decision neural network (DNN) for preference assessment

A new neural-network-based approach to assess the preference of a decision-maker (DM) for the multiple objective decision making (MODM) problem is presented in this paper. A new neural network structure with a "twin-topology" is introduced in this approach. We call this neural network a decision neural network (DNN). The characteristics of the DNN are discussed, and the training algorithm for DNN is presented as well. The DNN enables the decision-maker to make pairwise comparisons between different alternatives, and these comparison results are used as learning samples to train the DNN. The DNN is applicable for both accurate and inaccurate comparisons (results are given in approximate values or interval scales). The performance of the DNN is evaluated with several typical forms of utility functions. Results show that DNN is an effective and efficient way for modeling the preference of a decision-maker.     

A 1.2 GFLOPS neural network chip for high-speed neural networkservers

This paper describes a digital neural network chip for high-speed neural network servers. The chip employs single-instruction multiple-data stream (SIMD) architecture consisting of 12 floating-point processing units, a control unit, and a nonlinear function unit. At a 50 MHz clock frequency, the chip achieves a peak speed performance of 1.2 GFLOPS using 24-bit floating-point representation. Two schemes of expanding the network size enable neural tasks requiring over 1 million synapses to be executed. The average speed performances of typical neural network models are also discussed     

A vector neural network for emitter identification

This paper proposes a three-layer vector neural network (VNN) with a supervised learning algorithm suitable for signal classification in general, and for emitter identification (EID) in particular. The VNN can accept interval-value input data as well as scalar input data. The input features of the EID problems include the radio frequency, pulse width, and pulse repetition interval of a received emitter signal. Since the values of these features vary in interval ranges in accordance with a specific radar emitter, the VNN is proposed to process interval-value data in the EID problem. In the training phase, the interval values of the three features are presented to the input nodes of VNN. A new vector-type backpropagation learning algorithm is derived from an error function defined by the VNN's actual output and the desired output indicating the correct emitter type of the corresponding feature intervals. The algorithm can tune the weights of VNN optimally to approximate the nonlinear mapping between a given training set of feature intervals and the corresponding set of desired emitter types. After training, the VNN can be used to identify the sensed scalar-value features from a real-time received emitter signal. A number of simulations are presented to demonstrate the effectiveness and identification capability of VNN, including the two-EID problem and the multi-EID problem with/without additive noise. The simulated results show that the proposed algorithm cannot only accelerate the convergence speed, but it can help avoid getting stuck in bad local minima and achieve higher classification rate.     

A neural network architecture for preattentive vision

Recent results towards development of a neural network architecture for general-purpose preattentive vision are summarized. The architecture contains two parallel subsystems, the boundary contour system (BCS) and the feature contour system (FCS), which interact together to generate a representation of form-and-color-and-depth. Emergent boundary segmentation within the BCS and featural filling-in within the FCS are emphasized within a monocular setting. Applications to the analysis of boundaries, textures, and smooth surfaces are described, as is a model for invariant brightness perception under variable illumination conditions. The theory shows how suitably defined parallel and hierarchical interactions overcome computational uncertainties that necessarily exist at early processing stages. Some of the psychophysical and neurophysiological data supporting the theory's predictions are mentioned     

基于部分连接神经网络的序列数据分类算法研究

研究梳理了序列数据的定义,并且分析了5种类型的序列数据,结合局部连接神经网络的特点,研究了基于局部连接神经网络的序列数据的分类算法。使用该算法进行计算,学习与收敛速度较快,对于自适应建模与控制十分适用,利用方形基函数进行计算,在网络输出过程中注意只能利用方形函数来逼近光滑函数。由于序列数据分类运算在数据挖掘中存在巨大的优势,因此对序列数据算法的研究具有很高的理论与应用价值。     

时空道路网络连续最近邻查询的新方法

时空道路网络里连续最近邻查询在时空数据库查询的领域是一个新的重要课题，本文提出一个有效的方法（IE）以处理道路网络里的移动对象的k个最近邻居（k-NN）查询问题。     

A neural network compensator for uncertainties of roboticsmanipulators

A neural network controller for trajectory control of robotic manipulators that is used not to internalize the inverse dynamic model of the controlled object but to compensate only the uncertainties of the robotic manipulator is presented. Its performance is compared with that of the conventional adaptive scheme. The results show the ability of the neural network controller to adapt to unstructured effects. A learning method for the neural network compensator with true teaching signals is shown. The tracking error of the robotic manipulator was greatly reduced when this controller was used     

A NEURAL NETWORK HARDWARE MODEL FOR IMAGE THINNING

A new parallel thinning algorithm was proposed.This paper suggests a set of 5x5cloning templates of cellular neural network for image thinning.The principle of the templatesdesign and hardware model are proposed in the paper.Although the connected weights betweenthe neurons are asymmetric,it is shown that the network is stable,and it can be easily realized.     

A reconfigurable VLSI neural network

Due to the variety of architectures that need be considered while attempting solutions to various problems using neural networks, the implementation of a neural network with programmable topology and programmable weights has been undertaken. A new circuit block, the distributed neuron-synapse, has been used to implement a 1024 synapse reconfigurable network on a VLSI chip. In order to evaluate the performance of the VLSI chip, a complete test setup consisting of hardware for configuring the chip, programming the synaptic weights, presenting analog input vectors to the chip, and recording the outputs of the chip, has been built. Following the performance verification of each circuit block on the chip, various sample problems were solved. In each of the problems the synaptic weights were determined by training the neural network using a gradient-based learning algorithm which is incorporated in the experimental test setup. The results of this work indicate that reconfigurable neural networks built using distributed neuron synapses can be used to solve various problems efficiently     

A prediction-based neural network scheme for lossless data compression

This paper proposes a modified block-adaptive prediction-based neural network scheme for lossless data compression. A variety of neural network models from a selection of different network types, including feedforward, recurrent, and radial basis configurations are implemented with the scheme. The scheme is further expanded with combinations of popular lossless encoding algorithms. Simulation results are presented, taking characteristic features of the models, transmission issues, and practical considerations into account to determine optimized configuration, suitable training strategies, and implementation schemes. Estimations are used for comparisons of these characteristics with the existing schemes. It is also shown that the adaptations of the improvised scheme increases performance of even the classical predictors evaluated. In addition, the results obtained support that the total processing time of the two-stage scheme can, in certain cases, be faster than just using lossless encoders. Findings of the paper may be beneficial for future work, such as, in the hardware implementations of dedicated neural chips for lossless compression.     

A CASCADED MODEL OF NEURAL NETWORK FOR PATTERN RECOGNITION

A cascaded model of neural network and its learning algorithm suitable for opticalimplementation are proposed.Computer simulations have shown that this model may successfullybe applied to an error-tolerance pattern recognitions of multiple 3-D targets with arbitrary spatialorientations.     

A supervisory fuzzy neural network controller for slider-crank mechanism

A supervisory fuzzy neural network (FNN) controller is proposed to control a nonlinear slider-crank mechanism in this study. The control system is composed of a permanent magnet (PM) synchronous servo motor drive coupled with a slider-crank mechanism and a supervisory FNN position controller. The supervisory FNN controller comprises a sliding mode FNN controller and a supervisory controller. The sliding mode FNN controller combines the advantages of the sliding mode control with robust characteristics and the FNN with on-line learning ability. The supervisory controller is designed to stabilize the system states around a defined bound region. The theoretical and stability analyses of the supervisory FNN controller are discussed in detail. Simulation and experimental results are provided to show that the proposed control system is robust with regard to plant parameter variations and external load disturbance.     

A novel genetic-algorithm-based neural network for short-term load forecasting

This paper presents a neural network with a novel neuron model. In this model, the neuron has two activation functions and exhibits a node-to-node relationship in the hidden layer. This neural network provides better performance than a traditional feedforward neural network, and fewer hidden nodes are needed. The parameters of the proposed neural network are tuned by a genetic algorithm with arithmetic crossover and nonuniform mutation. Some applications are given to show the merits of the proposed neural network.     

A modular neural network vector predictor for predictive imagecoding

In this paper, we present a modular neural network vector predictor that improves the predictive component of a predictive vector quantization (PVQ) scheme. The proposed vector prediction technique consists of five dedicated predictors (experts), where each expert predictor is optimized for a particular class of input vectors. An input vector is classified into one of five classes, based on its directional variances. One expert predictor is optimized for stationary blocks, and each of the other four expert predictors are optimized to predict horizontal, vertical, 45 degrees , and 135 degrees diagonally oriented edge-blocks, respectively. An integrating unit is then used to select or combine the outputs of the experts in order to form the final output of the modular network. Therefore, no side information is transmitted to the receiver about the selected predictor or the integration of the predictors. Experimental results show that the proposed scheme gives an improvement of 1.7 dB over a single multilayer perceptron (MLP) predictor. Furthermore, if the information about the predictor selection is sent to the receiver, the improvement could be up to 3 dB over a single MLP predictor. The perceptual quality of the predicted images is also significantly improved.     

A neural network algorithm for sea ice edge classification

The NASA Scatterometer (NSCAT), launched in August 1995, is designed to measure wind vectors over ice-free oceans. To prevent contamination of the wind measurements, by the presence of sea ice, algorithms based on neural network technology have been developed to classify ice-free ocean surfaces. Neural networks trained using polarized alone and polarized plus multi-azimuth “look” Ku-band backscatter are described. Algorithm skill in locating the sea ice edge around Antarctica is experimentally evaluated using backscatter data from the Seasat-A Satellite Scatterometer that operated in 1978. Comparisons between the algorithms demonstrate a slight advantage of combined polarization and multi-look over using co-polarized backscatter alone. Classification skill is evaluated by comparisons with surface truth (sea ice maps), subjective ice classification, and independent over lapping scatterometer measurements (consecutive revolutions)     

A hybrid neural network model for predicting silicon content

The silicon content of the hot metal in the blast furnace ironmaking process normally reflects the thermal state of the furnace and affects the fuel rate. In this paper a hybrid neural network model is proposed to predict the silicon contentn steps ahead. A time-delay neural network, which has self-loops to represent dynamics, is adopted here. The learning procedure of this network has been divided into two states. A BP algorithm with forgetting factor is first introduced to find the appropriate structure of the network. The temporal difference (TD) method with forgetting factor is then used forn-step-ahead prediction. The results show that the method can perform satisfactoryn-step-ahead prediction and is suited for implementation.     

A fuzzy neural network controller for parallel-resonant ultrasonicmotor drive

A newly designed driving circuit for the traveling-wave-type ultrasonic motor (USM), which consists of a push-pull DC-DC power converter and a current-source two-phase parallel-resonant inverter, is presented in this study. Moreover, since the dynamic characteristics of the USM are difficult to obtain and the motor parameters are time varying, a fuzzy neural network (NN) controller is proposed to control the USM drive system. In the proposed controller, a fuzzy model-following controller is implemented to control the rotor position of the USM, and an online trained NN with variable learning rates is implemented to tune the output scaling factor of the fuzzy controller. To guarantee the convergence of tracking error, analytical methods based on a discrete-type Lyapunov function are proposed to determine the desired variable learning rates. From the experimental results, accurate tracking response can be obtained by the proposed controller, and the influences of parameter variations and external disturbances on the USM drive also can be reduced effectively     

A novel neural network traffic enforcement mechanism for ATMnetworks

ATM has been recommended by the CCITT as the transport vehicle for the future B-ISDN networks. In ATM-based networks, a set of user declared parameters that describes the traffic characteristics, is required for the connection acceptance control (CAC) and traffic enforcement (policing) mechanisms. At the call set-up phase, the CAC algorithm uses those parameters to make a call acceptance decision. During the call progress, the policing mechanism uses the same parameters to control the user's traffic within its declared values in order to protect the network's resources and avoid possible congestion problems. A novel policing mechanism using neural networks (NNs) is presented. This is based upon an accurate estimation of the probability density function (pdf) of the traffic via its count process and implemented using NNs. The pdf-based policing is made possible only by NNs because pdf policing requires complex calculations, in real-time, at very high speeds. The architecture of the policing mechanism is composed of two interconnected NNs. The first one is trained to learn the pdf of “ideal nonviolating” traffic, whereas the second is trained to capture the “actual” characteristics of the “actual” offered traffic during the progress of the call. The output of both NNs is compared. Consequently, an error signal is generated whenever the pdf of the offered traffic violates its “ideal” one. The error signal is then used to shape the traffic back to its original values     

A dynamic learning neural network for remote sensing applications

The neural network learning process is to adjust the network weights to adapt the selected training data. Based on the polynomial basis function (PBF) modeled neural network that is a modified multilayer perceptrons (MLP) network, a dynamic learning algorithm (DL) is proposed. The presented learning algorithm makes use of the Kalman filtering technique to update the network weights, in the sense that the stochastic characteristics of incoming data sets are implicitly incorporated into the network. The Kalman gains which represent the learning rates of the network weights updating are calculated by using the U-D factorization. By concatenating all of the network weights at each layer to form a long vector such that it can be updated without propagating back, the proposed algorithm improves the performance of convergence to which the backpropagation (BP) learning algorithm often suffers. Numerical illustrations are carried out using two categories of problems: multispectral imagery classification and surface parameters inversion. Results indicates the use of Kalman filtering algorithm not only substantially increases the convergence rate in the learning stage, but also enhances the separability for highly nonlinear boundaries problems, as compared to BP algorithm, suggesting that the proposed DL neural network provides a practical and potential tool for remote sensing applications     

A programmable analog neural network chip

A generic chip is implemented in CMOS to facilitate studying networks by building them in analog VLSI. By utilizing the well-known properties of charge storage and charge injection in a novel way, the authors have achieved a high enough level of complexity (>10³ weights and 10 bits of analog depth) to be interesting, in spite of the limitation of a modest 6.00×3.5-mm² die size required by a multiproject fabrication run. If the cell were optimized to represent fixed-weight networks by eliminating weight decay and bidirectional weight changes, the density could easily be increased by a factor of 2 with no loss in resolution. Once a weight change vector has been written to the RAM cells, charge transfers can be clocked at a rate of 2 MHz, corresponding to peak learning rates of 2×10⁹ weight changes/second and exceeding the throughput of `neural network accelerators' by two orders of magnitude