Stability and bifurcation in a neural network model with two delays

A simple neural network model with two delays is considered. Linear stability of the model is investigated by analyzing the associated characteristic transcendental equation. For the case without self-connection, it is found that the Hopf bifurcation occurs when the sum of the two delays varies and passes a sequence of critical values. The stability and direction of the Hopf bifurcation are determined by applying the normal form theory and the center manifold theorem. An example is given and numerical simulations are performed to illustrate the obtained results.     

Synchronization regimes in a map-based model neural network

The dynamical activity of a neural network model composed of electrically connected map-based neurons is investigated. After detailing the behavior of the isolated neuron for a wide parameter range, collective network states are depicted using the activity, spatial correlation and time phase distribution as measures. A detailed discussion on the stability of global and partial synchronization states is presented.     

Prediction of outdoor sound transmission loss with an artificial neural network

An artificial neural network is developed for rapid prediction of sound transmission loss (TL) during propagation outdoors. The network predicts TL for a nonturbulent atmosphere from inputs involving the source/receiver propagation geometry (height range: 0-5 m, horizontal separation distance: 100-900 m), source frequency (range: 20-200 Hz), ground properties, and atmospheric refractive profile characteristics. A parabolic equation (PE) code generates the training and test data sets for the network. To ensure that a minimal set of input parameters is used in the network training, a nondimensional version of the PE and accompanying boundary, initial, and atmospheric conditions is developed. A total of 10 independent, nondimensional input parameters are found to be necessary for the training. Approximately 27,000 random cases involving these 10 parameters are generated used to train networks with varying numbers of neurons. The root mean square (RMS) error between random test cases solved by the PE and corresponding neural network predictions was 2.42 dB when a sufficient number of neurons (about 44) are included in the hidden layer. Also, only 18% of the cases resulted in RMS errors that were greater than 2 dB.     

Composite MR image reconstruction and unaliasing for general trajectories using neural networks

In rapid parallel magnetic resonance imaging, the problem of image reconstruction is challenging. Here, a novel image reconstruction technique for data acquired along any general trajectory in neural network framework, called “Composite Reconstruction And Unaliasing using Neural Networks” (CRAUNN), is proposed. CRAUNN is based on the observation that the nature of aliasing remains unchanged whether the undersampled acquisition contains only low frequencies or includes high frequencies too. Here, the transformation needed to reconstruct the alias-free image from the aliased coil images is learnt, using acquisitions consisting of densely sampled low frequencies. Neural networks are made use of as machine learning tools to learn the transformation, in order to obtain the desired alias-free image for actual acquisitions containing sparsely sampled low as well as high frequencies. CRAUNN operates in the image domain and does not require explicit coil sensitivity estimation. It is also independent of the sampling trajectory used, and could be applied to arbitrary trajectories as well. As a pilot trial, the technique is first applied to Cartesian trajectory-sampled data. Experiments performed using radial and spiral trajectories on real and synthetic data, illustrate the performance of the method. The reconstruction errors depend on the acceleration factor as well as the sampling trajectory. It is found that higher acceleration factors can be obtained when radial trajectories are used. Comparisons against existing techniques are presented. CRAUNN has been found to perform on par with the state-of-the-art techniques. Acceleration factors of up to 4, 6 and 4 are achieved in Cartesian, radial and spiral cases, respectively.     

基于CMAC神经网络的能动磨盘智能控制实验研究

 针对能动磨盘面形控制系统的非线性和多变量特点,提出了基于CMAC神经网络的能动磨盘面形智能控制方法,以CMAC神经网络来映射磨盘面型和控制脉冲之间复杂的关系。为验证上述智能控制方法,搭建了由有效变形口径为420 mm能动磨盘和60路微位移阵列传感器组成的3单元能动磨盘面形检测实验平台,在该实验平台上进行了多组实验,利用微位移阵列传感器分别检测出能动磨盘在1单元、2单元和3单元驱动器作用下实验面形相对于理论面形的偏差,其中峰谷值分别为0.99,2.34和2.68 mm,均方根值分别为0.19,0.59和0.57 mm,实验结果验证了能动磨盘CMAC神经网络智能控制的可行性。     

Active noise control in a duct by an analog neural network circuit

Conventional active noise control (ANC) in ducts has been realized with digital signal processing. The physical size of the conventional ANC systems is usually large owing to the signal processing interval, and the cost of the system depends on the price of the digital signal processor (DSP). This paper proposes a new ANC system with an analog neural network circuit, which will process signals in short time periods without DSP. The proposed neural network circuit has a simple structure consisting of analog multipliers and an integrator, and we simulated the performance of the circuit by HSPICE. We also fabricated a circuit connected to a real duct and confirmed operation of the proposed ANC system.     

Model architecture for associative memory in a neural network of spiking neurons

A synaptic connectivity model is assembled on a spiking neuron network aiming to build up a dynamic pattern recognition system. The connection architecture includes gap junctions and both inhibitory and excitatory chemical synapses based on Hebb’s hypothesis. The network evolution resulting from external stimulus is sampled in a properly defined frequency space. Neurons’ responses to different current injections are mapped onto a subspace using Principal Component Analysis. Departing from the base attractor, related to a quiescent state, different external stimuli drive the network to different fixed points through specific trajectories in this subspace.     

Effect of delay on phase locking in a pulse coupled neural network

Using a slightly simplified version of the integrate and fire model of a neural network with delay, I study the stability of the phase-locked state dependent on the coupling between the neurons and especially on a delay time. The coupling between neurons may be arbitrary. It is shown that the phase-locked state becomes less stable with increasing delay and that relaxation oscillations occur. Received 28 December 1999 and Received in final form 13 June 2000     

Solder joint inspection based on neural network combined with genetic algorithm

To improve the performance of automatic optical inspection (AOI), a neural network combined with genetic algorithm for the diagnosis of solder joint defects on printed circuit boards (PCBs) assembled in surface mounting technology (SMT) is presented. Six types of solder joint have been classified in respect to the reality in the manufacture. The images of solder joint under test are acquired and 14 features are extracted as input features for the classification. The neural network is easily become over-fitting because these input features are not independent of each other, so the genetic algorithm is introduced to select and remove redundant input features. The experimental results have proved that the neural network combined with genetic algorithm reduced the number of input feature and had a satisfying recognition rate.     

An effective hybrid V2V/V2I transmission latency method based on LSTM neural network

We propose an effective hybrid vehicle-to-vehicle/vehicle-to-infrastructure (V2V/V2I) transmission latency method based on a long short-term memory (LSTM) neural network to address transmission latency in the internet of vehicles. First, a traffic model is established, and the LSTM artificial neural network is used to predict the vehicle arrival rate in the road section. Second, the vehicle arrival rate function is used to construct an objective function, i.e., the problem of minimizing system transmission the overall latency. The hybrid V2V/V2I transmission method determines the communication transmission mode of the vehicles to minimize transmission latency. The simulation results show that the overall transmission latency is substantially lower for the hybrid V2V/V2I transmission method than the pure V2I transmission method with the transmission packet size and vehicle speed varying.     

Clustering in neural networks with heterogeneous and asymmetrical coupling strengths

The existence and stability of phase-clustered states have been studied previously in networks of weakly coupled oscillators with uniform coupling strengths [Physica D 63 (1993) 424]. However, several studies have shown that if the coupling is uniform and repulsive, it is hard to obtain stable phase-clustered states in networks of realistic neural oscillators when noise is present [Neural Comput. 7 (1995) 307; Phys. Rev. E 57 (1998) 2150]. This problem was avoided by introducing heterogeneity in the distribution of coupling strengths [J. Phys. Soc. Jpn. 72 (2003) 443]. It has been shown that heterogeneous coupling strengths make the occurrence of stable clustered states possible in small networks of repulsively coupled neural oscillators of all kinds [J. Comput. Neurosci. 14 (2003) 139; SIAM J. Appl. Math., submitted for publication]. The present work extends these results to large networks of N identical neurons that are globally coupled with heterogeneous and asymmetrical coupling strengths. Conditions for the existence and stability of a state of n synchronized clusters at evenly distributed phases, called the state of n splay-phase clusters, are derived. Clusters of different sizes, i.e. containing different numbers of neurons, are studied. The existence of such a state is guaranteed if the strength of the coupling originating from one neuron to other neurons is inversely proportional to the size of the cluster to which it belongs. This condition is called the rule of inverse cluster-size. At the state of n splay-phase clusters, the N-neuron network behaves like a network of n “big neurons”. Stability of this state is determined by n eigenvalues of which only one determines the stability of intra-cluster phase differences. The remaining n−1 conditions determine the stability of inter-cluster phase differences, but only n_h=(n− mod (n,2))/2 of them have distinct real parts due to symmetry. Heterogeneous coupling makes the stability conditions depend on coupling strengths. This analysis not only reveals how clustered states occur in more general kinds of networks, but also illustrates how the stability of clustered states can be achieved in networks of repulsively coupled neural oscillators. Results on clustered states with phases that are not evenly distributed in the phase space are also presented. Potential applications of these results are discussed.     

水下高分辨率声图中小目标的深度网络分类方法

针对声成像数据缺少条件下的水下沉底小目标分类问题,提出一种深度网络分类算法。首先,采用高斯混合模型对声影区统计特性进行建模并提取声图阴影,在此基础上构建仿真数据集和真实数据集。将仿真数据集输入卷积神经网络进行训练,保留其特征提取部分,用于对真实数据集进行特征提取.重建网络分类部分并采用真实数据集的特征向量进行训练。结果表明,所提出的方法分类正确率可达88.24%,与6种对照方法相比平均分类正确率分别提升8.67%,20.47%,19.78%,11.59%,9.01%,11.58%。验证了所提出方法在小样本条件下具有较好对水下沉底小目标的分类能力。其学习曲线收敛到96.25%,仅比验证曲线高5.14%,说明在一定程度上缓解了过拟合问题。将改进的卷积神经网络应用于融合分类器,通过与逻辑回归分类器、支持向量机对目标进行分类并融合决策,正确率为93.33%,可进一步提高算法的正确率和稳定性.     

不确定时滞混沌系统的自适应动态神经网络控制

研究了一类具有不确定时滞的非自治混沌系统的控制问题. 通过结合Lyapunov-Krasovskii函数和Lyapunov函数设计参数可调的不确定时滞补偿器,使得反馈控制输入信号不受时延的影响;同时引入动态结构自适应神经网络,以消除系统的不确定性,其隐层神经元的个数可以随着逼近误差的增大而自适应增加,改善了逼近速度与网络复杂度的关系;最后,用Duffing混沌系统的控制仿真示例表明该方法的有效性. 关键词： 混沌系统 自适应控制 不确定时滞 动态结构神经网络     

Super-resolution imaging in digital holography by using dynamic grating with a spatial light modulator

A super-resolution imaging method using dynamic grating based on liquid-crystal spatial light modulator (SLM) is developed to improve the resolution of a digital holographic system. The one-dimensional amplitude cosine grating is loaded on the SLM, which is placed between the object and hologram plane in order to collect more high-frequency components towards CCD plane. The point spread function of the system is given to confirm the separation condition of reconstructed images for multiple diffraction orders. The simulation and experiments are carried out for a standard resolution test target as a sample, which confirms that the imaging resolution is improved from 55.7 μm to 31.3 μm compared with traditional lensless Fourier transform digital holography. The unique advantage of the proposed method is that the period of the grating can be programmably adjusted according to the separation condition.     

Ionospheric vertical total electron content prediction model in low-latitude regions based on long short-term memory neural network

Ionosphere delay is one of the main sources of noise affecting global navigation satellite systems, operation of radio detection and ranging systems and very-long-baseline-interferometry. One of the most important and common methods to reduce this phase delay is to establish accurate nowcasting and forecasting ionospheric total electron content models. For forecasting models, compared to mid-to-high latitudes, at low latitudes, an active ionosphere leads to extreme differences between long-term prediction models and the actual state of the ionosphere. To solve the problem of low accuracy for long-term prediction models at low latitudes, this article provides a low-latitude, long-term ionospheric prediction model based on a multi-input-multi-output, long-short-term memory neural network. To verify the feasibility of the model, we first made predictions of the vertical total electron content data 24 and 48 hours in advance for each day of July 2020 and then compared both the predictions corresponding to a given day, for all days. Furthermore, in the model modification part, we selected historical data from June 2020 for the validation set, determined a large offset from the results that were predicted to be active, and used the ratio of the mean absolute error of the detected results to that of the predicted results as a correction coefficient to modify our multi-input-multi-output long short-term memory model. The average root mean square error of the 24-hour-advance predictions of our modified model was 4.4 TECU, which was lower and better than 5.1 TECU of the multi-input-multi-output, long short-term memory model and 5.9 TECU of the IRI-2016 model.     

基于径向基函数神经网络的混沌时间序列相空间重构双参数联合估计

鉴于径向基函数(RBF)神经网络模型在非线性预测方面的优良性能, 提出了利用该预测模型对混沌时间序列相空间重构的两个关键参数——延迟时间和嵌入维数进行联合估计的方法, 并以客观的评价指标为依据给出其最优估计值. 以Lorenz系统为例进行数值分析, 得到RBF单步及多步预测模型中嵌入维数和延迟时间的最佳参数估计值, 并在原模型中对估计值进行校验. 结果表明, 该方法可以有效地估计出嵌入维数和延迟时间, 从而显著提高预测精度.     

A different approach to gain computation in laser diodes with respect to different number of quantum-wells

This study presents a different approach for the modelling of optical gain in laser diodes as a function of quantum-well (QW) number based on Artificial Neural Networks (ANNs). Different learning algorithms with different network configurations are tried and tested in order to minimize the rms errors in terms of the ANN structure, number of layers, and number of neurons in each layer. The optical gain results obtained by using this method are in very good agreement with the experimental results reported elsewhere.     

Detecting community structure using label propagation with consensus weight in complex network

Community detection is a fundamental work to analyse the structural and functional properties of complex networks.The label propagation algorithm(LPA) is a near linear time algorithm to find a good community structure. Despite various ubsequent advances, an important issue of this algorithm has not yet been properly addressed. Random update orders within the algorithm severely hamper the stability of the identified community structure. In this paper, we executed the asic label propagation algorithm on networks multiple times, to obtain a set of consensus partitions. Based on these onsensus partitions, we created a consensus weighted graph. In this consensus weighted graph, the weight value of the dge was the proportion value that the number of node pairs allocated in the same cluster was divided by the total number f partitions. Then, we introduced consensus weight to indicate the direction of label propagation. In label update steps,y computing the mixing value of consensus weight and label frequency, a node adopted the label which has the maximum mixing value instead of the most frequent one. For extending to different networks, we introduced a proportion parameter o adjust the proportion of consensus weight and label frequency in computing mixing value. Finally, we proposed an pproach named the label propagation algorithm with consensus weight(LPAcw), and the experimental results showed that he LPAcw could enhance considerably both the stability and the accuracy of community partitions.     

Detection of material property errors in handbooks and databases using artificial neural networks with hidden correlations

The authors have discovered a systematic, intelligent and potentially automatic method to detect errors in handbooks and stop their transmission using unrecognised relationships between materials properties. The scientific community relies on the veracity of scientific data in handbooks and databases, some of which have a long pedigree covering several decades. Although various outlier-detection procedures are employed to detect and, where appropriate, remove contaminated data, errors, which had not been discovered by established methods, were easily detected by our artificial neural network in tables of properties of the elements. We started using neural networks to discover unrecognised relationships between materials properties and quickly found that they were very good at finding inconsistencies in groups of data. They reveal variations from 10 to 900% in tables of property data for the elements and point out those that are most probably correct. Compared with the statistical method adopted by Ashby and co-workers [Proc. R. Soc. Lond. Ser. A 454 (1998) p. 1301, 1323], this method locates more inconsistencies and could be embedded in database software for automatic self-checking. We anticipate that our suggestion will be a starting point to deal with this basic problem that affects researchers in every field. The authors believe it may eventually moderate the current expectation that data field error rates will persist at between 1 and 5%.     

Quality of Service based resource allocation in D2D enabled 5G-CNs with network slicing

With the rapid new advancements in technology, there is an enormous increase in devices and their versatile need for services. Fifth-generation (5G) cellular networks (5G-CNs) with network slicing (NS) have emerged as a necessity for future mobile communication. The available network is partitioned logically into multiple virtual networks to provide an enormous range of users’ specific services. Efficient resource allocation methods are critical to delivering the customers with their required Quality of Service (QoS) priorities. In this work, we have investigated a QoS based resource allocation (RA) scheme considering two types of 5G slices with different service requirements; (1) enhanced Mobile Broadband (eMBB) slice that requires a very high data rate and (2) massive Machine Type Communication (mMTC) slice that requires extremely low latency. We investigated the device-to-device (D2D) enabled 5G-CN model with NS to assign resources to users based on their QoS needs while considering the cellular and D2D user’s data rate requirements. We have proposed a Distributed Algorithm (DA) with edge computation to solve the optimization problem, which is novel as edge routers will solve the problem locally using the augmented Lagrange method. They then send this information to the central server to find the global optimum solution utilizing a consensus algorithm. Simulation analysis proves that this scheme is efficient as it assigns resources based on their QoS requirements. This scheme is excellent in reducing the central load and computational time.