Network restoration using recurrent neural networks

In this article, a method is proposed for network restoration using a centralized, static restoration after failure, where the restoration initiated at the local node or at the source uses a hybrid strategy. © 1998 John Wiley & Sons, Ltd.     

Interpolation of LSP coefficients using recurrent neural networks

To reduce the bit rate in very low bit rate speech coding, recurrent neural networks (RNNs) were applied to interpolate LSP coefficients, and it was possible to increase the duration of interpolation to ~100 ms. Simulation of coding was carried out, and the proposed method outperformed the other interpolation methods below a bit rate of 300 bit/s     

Bidirectional recurrent neural networks

In the first part of this paper, a regular recurrent neural network (RNN) is extended to a bidirectional recurrent neural network (BRNN). The BRNN can be trained without the limitation of using input information just up to a preset future frame. This is accomplished by training it simultaneously in positive and negative time direction. Structure and training procedure of the proposed network are explained. In regression and classification experiments on artificial data, the proposed structure gives better results than other approaches. For real data, classification experiments for phonemes from the TIMIT database show the same tendency. In the second part of this paper, it is shown how the proposed bidirectional structure can be easily modified to allow efficient estimation of the conditional posterior probability of complete symbol sequences without making any explicit assumption about the shape of the distribution. For this part, experiments on real data are reported     

Prediction of MPEG-coded video source traffic using recurrent neural networks

Predicting traffic generated by multimedia sources is needed for effective dynamic bandwidth allocation and for multimedia quality-of-service (QoS) control strategies implemented at the network edges. The time-series representing frame or visual object plane (VOP) sizes of an MPEG-coded stream is extremely noisy, and it has very long-range time dependencies. This paper provides an approach for developing MPEG-coded real-time video traffic predictors for use in single-step (SS) and multistep (MS) prediction horizons. The designed SS predictor consists of one recurrent network for I-VOPs and two feedforward networks for P- and B-VOPs, respectively. These are used for single-frame-ahead prediction. A moving average of the frame or VOP sizes time-series is generated from the individual frame sizes and used for both SS and MS prediction. The resulting MS predictor is based on recurrent networks, and it is used to perform two-step-ahead and four-step-ahead prediction, corresponding to multistep prediction horizons of 1 and 2 s, respectively. All of the predictors are designed using a segment of a single MPEG-4 video stream, and they are tested for accuracy on complete video streams with a variety of quantization levels, coded with both MPEG-1 and MPEG-4. Comparisons with SS prediction results of MPEG-1 coded video traces from the recent literature are presented. No similar results are available for prediction of MPEG-4 coded video traces and for MS prediction. These are considered unique contributions of this research.     

一种基于脉冲耦合神经网络和图像熵的自动图像分割方法

90年代发展形成的脉冲耦合神经网络（PCNN）模型特别适合于图像分割、边缘提取等方面的应用研究，但众所周知，PCNN模型图像分割效果不但取决于PCNN模型中各个参数的合理选择，而且同时还取决于循环迭代次数的确定选择准则，通常循环迭代次数N的选择通过人工交互方式来确定。正因为如此选择合适的准则来确定N是PCNN图像分割的关键，但目前还没有文献提出一个合适的准则来解决这个问题。本文结合图像统计特性和PCNN参数模型提出了熵值最大准则。该准则实现了PCNN神经网络的自动图像分割。对于PCNN的理论研究和实际应用具有非常重要的现实意义。     

Baltic Sea ice SAR segmentation and classification using modified pulse-coupled neural networks

A method for segmentation and classification of Baltic Sea ice synthetic aperture radar (SAR) images, based on pulse-coupled neural networks (PCNNs), is presented. Also, automated training, which is based on decomposing the total pixel value distribution into a mixture of class distributions, is presented and discussed. The algorithms have been trained and tested using logarithmic scale Radarsat-1 ScanSAR Wide mode images over the Baltic Sea ice. Before the decomposition into mixture of class distributions, an incidence angle correction, specifically designed for these Baltic Sea ice SAR images, is applied. Because the data distributions in the uniform areas of these images are very close to Gaussian distributions, the data are decomposed into a mixture of Gaussian distributions, using the Expectation-Maximazation algorithm. Only uniform image areas are used in the decomposition phase. The mixture of distributions is compared to the distributions of the Baltic Sea ice classes, based on earlier scatterometer measurements and visual video interpretations of the sea ice classes. The parameter values for the PCNN segmentation are defined based on this mixture of distributions. The PCNN segmentation results are also compared to the operational sea ice information of digitized ice charts and to visual interpretation of the sea ice class.     

采用门控循环神经网络估计锂离子电池健康状态

锂离子电池健康状态（State of Health，SOH）描述了电池当前老化程度，对于提前对电池的故障及失控做出预警避免电池的不安全行为具有重要意义。其估计难点在于难以确定数量合适、相关性高的估计输入以及设计合适的估计算法。通过对现有电池老化数据集的研究发现，电池充电过程中电压曲线数据相对稳定，且随着电池的老化出现规律性变化。因此，文中直接采用充电过程中电压数据作为估计SOH的输入，并在数据驱动的框架下，提出了一种基于门控循环神经网络（Recurrent Neural Networks with Gated Recurrent Unit, GRU-RNN）的锂电池SOH估计方法。该方法能够挖掘出一维电压数据中的时序特征和SOH之间的映射规律。在两个公开的电池老化数据集上的实验结果表明，提出的方法达到了1.25%的均方绝对误差和低于5.62%的最大误差，在估计精度上达到现有技术发展水平。     

Narrow-band interference suppression in spread-spectrum CDMAcommunications using pipelined recurrent neural networks

This paper investigates the application of pipelined recurrent neural networks (PRNN's) to the narrow-band interference (NBI) suppression over spread-spectrum (SS) code-division multiple-access (CDMA) channels in the presence of additive white Gaussian noise (AWGN) plus non-Gaussian observation noise. Optimal detectors and receivers for such channels are no longer linear. A PRNN that consists of a number of simpler small-scale recurrent neural network (RNN) modules with less computational complexity is conducted to introduce best nonlinear approximation capability into the minimum mean-squared error nonlinear predictor model in order to accurately predict the NBI signal based on adaptive learning for each module from previous non-Gaussian observations. Once the prediction of the NBI signal is obtained, a resulting signal is computed by subtracting the estimate from the received signal. Thus, the effect of the NBI can be reduced. Moreover, since those modules of a PRNN can be performed simultaneously in a pipelined parallelism fashion, this would lead to a significant improvement in its total computational efficiency. Simulation results show that PRNN-based NBI rejection provides a superior signal-to-noise ratio (SNR) improvement relative to the conventional adaptive nonlinear approximate conditional mean (ACM) filters, especially when the channel statistics and exact number of CDMA users are not known to those receivers     

Application of neural networks to AVHRR cloud segmentation

The application of neural networks to cloud screening of AVHRR data over the ocean is investigated. Two approaches are considered, interactive cloud screening and automated cloud screening. In interactive cloud screening a neural network is trained on a set of data points which are interactively selected from the image to be screened. Because the data variability is limited within a single image, a very simple neural network topology is sufficient to generate an effective cloud screen. Consequently, network training is very quick and only a few training samples are required. In automated cloud screening, where a general network is designed to handle all images, the data variability can be significant and the resulting neural network topology is more complex. The latitudinal, seasonal and spatial dependence of cloud screening large AVHRR data sets is studied using an extensive data set spanning 7 years. A neural network and associated feature set are designed to cloud screen this data set. The sensitivity of the thermal infrared bands to high atmospheric water vapor concentration was found to limit the accuracy of cloud screening methods which rely solely on data from these channels. These limitations are removed when the visible channel data is used in combination with the thermal infrared data. A post processing algorithm is developed to improve the cloud screening results of the network in the presence of high atmospheric water vapor concentration. Post processing also is effective in identifying pixels contaminated by subpixel clouds and/or amplifier hysteresis effects at cloud-ocean boundaries. The neural network, when combined with the post processing algorithm, produces accurate cloud screens for the large, regionally distributed AVHRR data set     

Optimal nonlinear adaptive prediction and modeling of MPEG video inATM networks using pipelined recurrent neural networks

This paper investigates the application of a pipelined recurrent neural network (PRNN) to the adaptive traffic prediction of MPEG video signal via dynamic ATM networks. The traffic signal of each picture type (I, P, and B) of MPEG video is characterized by a general nonlinear autoregressive moving average (NARMA) process. Moreover, a minimum mean-squared error predictor based on the NARMA model is developed to provide the best prediction for the video traffic signal. However, the explicit functional expression of the best mean-squared error predictor is actually unknown. To tackle this difficulty, a PRNN that consists of a number of simpler small-scale recurrent neural network (RNN) modules with less computational complexity is conducted to introduce the best nonlinear approximation capability into the minimum mean-squared error predictor model in order to accurately predict the future behavior of MPEG video traffic in a relatively short time period based on adaptive learning for each module from previous measurement data, in order to provide faster and more accurate control action to avoid the effects of excessive load situation. Since those modules of PRNN can be performed simultaneously in a pipelined parallelism fashion, this would lead to a significant improvement in the total computational efficiency of PRNN. In order to further improve the convergence performance of the adaptive algorithm for PRNN, a learning-rate annealing schedule is proposed to accelerate the adaptive learning process. Another advantage of the PRNN-based predictor is its generalization from learning that is useful for learning a dynamic environment for MPEG video traffic prediction in ATM networks where observations may be incomplete, delayed, or partially available. The PRNN-based predictor presented in this paper is shown to be promising and practically feasible in obtaining the best adaptive prediction of real-time MPEG video traffic     

基于两种神经网络的SAR图像多尺度分割与分析

基于多尺度回归技术和神经网络提出两种合成孔径雷达(SAR)图像分割的新方法.首先利用多尺度自回归模型(MAR)来描述SAR图像不同尺度间的统计相依性,以此提取SAR图像的多尺度统计特征;然后分别构造自组织特征映射网络和概率神经网络,并利用统计特征作为输入训练两种网络,实现SAR图像的分割.最后通过实验对这两种方法以及其他方法之间进行比较、分析,结果表明本文提出的两种方法的实验结果比较理想.     

递归神经网络自适应均衡抗突发干扰研究

短波信道的突发干扰对判决反馈递归神经网络自适应均衡器有破坏作用并能导致均衡器失效,从而影响整个通信系统的连续性。本文对突发干扰时判决反馈递归神经网络自适应均衡器特性进行了分析,提出了2种改进算法。仿真结果证明了这2种改进算法的有效性。     

ANALYSIS AND IMPROVEMENT OF RECURRENT CORRELATION NEURAL NETWORKS

This paper analyzes the relationship between capacity and dynamics in recurrent correlation neural network, and points out that in some conditions the recurrent correlation neural network has high memory capacity. Then this paper presents several methods for improving the performance.     

Nonlinear adaptive aggressive control using recurrent neural networks for a small scale helicopter

This paper presents a nonlinear adaptive aggressive controller to provide the small scale helicopter with full authority of a variety of flight conditions. Adaptive backstepping technique is employed to systematically synthesize the proposed controller with the online parameter adaptation rule to the vehicle mass variations and with the recurrent neural network (RNN) approximation to the coupling effect between the force and moment controls. This single and systematic design methodology is shown to achieve the semi-global ultimate boundedness of the closed-loop helicopter dynamics and accommodate the aggressive control of flight maneuvers from hovering to trajectory tracking. The high-fidelity and well-validated nonlinear model of a small scale helicopter incorporating with unmodeled dynamics and measurement uncertainties is adopted in the numerical simulations. The performance and merits of the proposed controller are exemplified by conducting three simulation scenarios including the slalom maneuver described in the ADS33.     

Fast adaptive digital equalization by recurrent neural networks

Neural networks (NNs) have been extensively applied to many signal processing problems. In particular, due to their capacity to form complex decision regions, NNs have been successfully used in adaptive equalization of digital communication channels. The mean square error (MSE) criterion, which is usually adopted in neural learning, is not directly related to the minimization of the classification error, i.e., bit error rate (BER), which is of interest in channel equalization. Moreover, common gradient-based learning techniques are often characterized by slow speed of convergence and numerical ill conditioning. In this paper, we introduce a novel approach to learning in recurrent neural networks (RNNs) that exploits the principle of discriminative learning, minimizing an error functional that is a direct measure of the classification error. The proposed method extends to RNNs a technique applied with success to fast learning of feedforward NNs and is based on the descent of the error functional in the space of the linear combinations of the neurons (the neuron space); its main features are higher speed of convergence and better numerical conditioning w.r.t. gradient-based approaches, whereas numerical stability is assured by the use of robust least squares solvers. Experiments regarding the equalization of PAM signals in different transmission channels are described, which demonstrate the effectiveness of the proposed approach     

Optimization neural networks for the segmentation of magnetic resonance images

The application of the Hopfield neural network for the multispectral unsupervised classification of MR images is reported. Winner-take-all neurons were used to obtain a crisp classification map using proton density-weighted and T(2)-weighted images in the head. The preliminary studies indicate that the number of iterations needed to reach ;good' solutions was nearly constant with the number of clusters chosen for the problem.     

基于神经网络的几何模型区域分割算法

三维模型简化是近年来计算机图形学中的一个研究热点,现有的简化算法多从全局出发,对几何模型的各个部位统一进行简化,因此模型简化后大量的细节特征丢失.针对三维模型简化中保留细节特征的需要,提出了一种基于自组织特征映射神经网络的三维模型区域分割算法.首先计算三维几何模型中每一顶点的特征向量,然后利用该向量作为自组织特征映射神经网络的输入模式实现对三维模型的聚类分割,最后采取提出的相关性最大准则对过分割区域进行合并,得到最终分割结果.实验表明,该方法能有效地分割出模型的细节区域,满足三维模型简化中保留细节特征的需要.     

Global convergence of Lotka-Volterra recurrent neural networks with delays

Recurrent neural networks of the Lotka-Volterra model have been proven to possess characteristics which are desirable in some neural computations. A clear understanding of the dynamical properties of a recurrent neural network is necessary for efficient applications of the network. This paper studies the global convergence of general Lotka-Volterra recurrent neural networks with variable delays. The contributions of this paper are: 1) sufficient conditions are established for lower positive boundedness of the networks; 2) global exponential stability conditions are obtained for the networks. These conditions are totally independent of the variable delays which are therefore allowed to be uncertain; 3) novel Lyapunov functionals are constructed to establish delays dependent conditions for global asymptotic stability, and 4) simulation results and examples are provided to supplement and illustrate the theoretical contributions presented.     

Image interpolation using neural networks

This work presents an image interpolation method based on a multilayer perceptron. The method is tested in noise-free as well as noisy line doubling and image expansion problems. Two adaptive algorithms are compared. Results show that the proposed method improves image interpolation.     

Approximation of dynamical time-variant systems by continuous-time recurrent neural networks

This paper studies the approximation ability of continuous-time recurrent neural networks to dynamical time-variant systems. It proves that any finite time trajectory of a given dynamical time-variant system can be approximated by the internal state of a continuous-time recurrent neural network. Given several special forms of dynamical time-variant systems or trajectories, this paper shows that they can all be approximately realized by the internal state of a simple recurrent neural network.