Synthesization of high-capacity auto-associative memories using complex-valued neural networks

In this paper, a novel design procedure is proposed for synthesizing high-capacity auto-associative memories based on complex-valued neural networks with real-imaginary-type activation functions and constant delays. Stability criteria dependent on external inputs of neural networks are derived. The designed networks can retrieve the stored patterns by external inputs rather than initial conditions. The derivation can memorize the desired patterns with lower-dimensional neural networks than real-valued neural networks, and eliminate spurious equilibria of complex-valued neural networks.One numerical example is provided to show the effectiveness and superiority of the presented results.     

Influence of Blurred Ways on Pattern Recognition of a Scale-Free Hopfield Neural Network

We investigate the influence of blurred ways on pattern recognition of a Barabasi-Albert scale-free Hopfield neural network （SFHN） with a small amount of errors. Pattern recognition is an important function of information processing in brain. Due to heterogeneous degree of scale-free network, different blurred ways have different influences on pattern recognition with same errors. Simulation shows that among partial recognition, the larger loading ratio （the number of patterns to average degree P/ （k） ） is, the smaller the overlap of SFHN is. The influence of directed （large） way is largest and the directed （small） way is smallest while random way is intermediate between them. Under the ratio of the numbers of stored patterns to the size of the network PIN is less than O. 1 conditions, there are three families curves of the overlap corresponding to directed （small）, random and directed （large） blurred ways of patterns and these curves are not associated with the size of network and the number of patterns. This phenomenon only occurs in the SFHN. These conclusions are benefit for understanding the relation between neural network structure and brain function.     

Structured information in sparse-code metric neural networks

Sparse-code networks have retrieval abilities which are strongly dependent on the firing threshold for the neurons. If the connections are spatially uniform, the macroscopic properties of the network can be measured by the overlap between neurons and learned patterns, and by the global activity. However, for nonuniform networks, for instance small-world networks, the neurons can retrieve fragments of patterns without performing global retrieval. Local overlaps are needed to describe the network. We characterize the structure type of the neural states using a parameter that is related to fluctuations of the local overlaps, with distinction between bump and block phases. Simulation of neural dynamics shows a competition between localized (bump), structured (block) and global retrieval. When the network topology randomness increases, the phase-diagram shows a transition from local to global retrieval. Furthermore, the local phase splits into a bump phase for low activity and a block phase for high activity. A theoretical approach solves the asymptotic limit of the model, and confirms the simulation results which predicts the change of stability from bumps to blocks when the storage ratio increases.     

混沌神经网络的延时反馈控制研究

针对混沌神经网络，提出了一种改进的延时反馈控制方法. 利用该方法，当延时参数τ为奇数时，被控神经网络收敛于记忆模式以及它的反相模式的2周期上. 若选取不同的延时参数，被控网络则收敛于不同的周期态上. 关键词： 控制混沌 延时反馈控制 混沌神经网络     

Real-time determination of fringe pattern frequencies: An application to pressure measurement

Retrieving information in real time from fringe patterns is a topic of a great deal of interest in scientific and engineering applications of optical methods. This paper presents a method for fringe frequency determination based on the capability of neural networks to recognize signals that are similar but not identical to signals used to train the neural network. Sampled patterns are generated by calibration and stored in memory. Incoming patterns are analyzed by a back-propagation neural network at the speed of the recording device, a CCD camera. This method of information retrieval is utilized to measure pressures on a boundary layer flow. The sensor combines optics and electronics to analyze dynamic pressure distributions and to feed information to a control system that is capable to preserve the stability of the flow.     

不确定混沌系统的回归神经网络同步

将高阶连接的神经元融合到分布式回归神经网络，研究了出现非模型动态性时不确定混沌系统的辨识和同步问题。采用李雅谱诺夫稳定理论对高阶神经网络回归模型的权值进行学习更新，同时，获取整个系统稳定特性的分析结果，而且通过李雅谱诺夫方法设计出消除不确定混沌系统的同步误差的自适应控制律。最后将所提出的方法应用到不确定Rossler混沌系统的建模与同步     

分区适应截值模型及其在交通标志识别中的应用

针对Ｈｏｐｆｉｅｌｄ网络模型在存储模式不满足０和１状态的均匀分布及数目对等的条件下存储容量及寻址能力下降的缺点，提出并用光束方向编码光学实现了三值（１，０，－１）互连的分区适应截值模型，并把这一模型应用到交通标志的识别中，结果表明该模型及光学系统有很好的稳定性。     

神经系统中随机和混沌感知信号的联想记忆与分割

为了模拟人与动物感知信息的真实环境，以脉动神经元节点组成神经元网络，研究在随机刺激和混沌刺激等极端条件下的记忆模式存储与时间分割问题.研究表明：网络对于若干种模式的叠加输入，能够以一部分神经元同步发放的形式在时间域上分割出每一模式. 如果输入模式是缺损的，系统能够把它们恢复到原型，即具有联想记忆功能.通过调节耦合强度和噪声强度等参数使得网络在中等强度噪声达到最优的时间分割，与广泛讨论的随机共振现象一致. 关键词： 神经网络 空时模式 联想记忆 随机共振     

Coarse-Grained Pruning of Neural Network Models Based on Blocky Sparse Structure

Deep neural networks may achieve excellent performance in many research fields. However, many deep neural network models are over-parameterized. The computation of weight matrices often consumes a lot of time, which requires plenty of computing resources. In order to solve these problems, a novel block-based division method and a special coarse-grained block pruning strategy are proposed in this paper to simplify and compress the fully connected structure, and the pruned weight matrices with a blocky structure are then stored in the format of Block Sparse Row (BSR) to accelerate the calculation of the weight matrices. First, the weight matrices are divided into square sub-blocks based on spatial aggregation. Second, a coarse-grained block pruning procedure is utilized to scale down the model parameters. Finally, the BSR storage format, which is much more friendly to block sparse matrix storage and computation, is employed to store these pruned dense weight blocks to speed up the calculation. In the following experiments on MNIST and Fashion-MNIST datasets, the trend of accuracies with different pruning granularities and different sparsity is explored in order to analyze our method. The experimental results show that our coarse-grained block pruning method can compress the network and can reduce the computational cost without greatly degrading the classification accuracy. The experiment on the CIFAR-10 dataset shows that our block pruning strategy can combine well with the convolutional networks.     

粉末涂料电脑配色的人工神经网络模型

提出一种基于多层BP人工神经网络的粉末涂料配方预测模型;用BP算法人工神经网络建立粉末涂料反射样品的标准色度参数与配方浓度参数之间的映射关系。把人工神经网络的配方预测模型应用到典型的粉末涂料样品的测配色实验过程中。实验结果表明,基于多隐层BP网的模型可以实现粉末涂料样品的配方浓度空间与标准三刺激值颜色空间的相互映射,对64个节点的平均训练精度达到了1个CIELAB色差单位。     

Pair-correlated patterns in Hopfield model of neural networks

We study the retrieval properties of the Hopfield model of neural networks when the memorized patterns are statistically correlated in pairs. There is a finite correlationk between the memories of each pair, but memories of different pairs are uncorrelated. The analysis is restricted to the case of an arbitrary but finite number of memories in the thermodynamic limit. We find that there are two retrieval regimes: for 0<T<(1–k) the system recognizes the stored patterns and for (1–k)<T<(1+k) the system is able to recognize pairs, but it is not able to distinguish between its two patterns.     

时滞双向联想记忆神经网络的全局稳定性

通过构造一个合适的Lyapunov泛函及应用不等式的分析技巧研究了具有时滞的双向联想记忆 神经网络的平衡点的全局稳定性问题-在对神经元激励函数较宽松的假设条件下（可以不满 足Lipschitz条件），获得了一个新的保证全局渐近稳定性的判定准则-结果可应用于包含非 Lipschitz的一类更加广泛的神经元激励函数的神经网络的设计中- 关键词： Lyapunov泛函 时滞 双向联想记忆神经网络 全局渐近稳定性     

基于混沌神经网络的盲检测改进新算法

针对Hopfield神经网络的多起点问题,提出了一种新的基于混沌神经网络的盲信号检测算法,实现了二进制移相键控信号盲检测.据此进一步提出双sigmoid混沌神经网络模型,构造了新的能量函数,且证明了该模型的稳定性,并对网络参数进行配置.仿真实验表明:混沌神经网络能够避免局部极小点且具备较强的抗噪性能,双sigmoid混沌神经网络则继承了其所有的优点,且其收敛速度更快,仅需更短的接收数据即可到达全局真实平衡点,从而降低了算法的计算复杂度,减少了运行时间.     

Phase transition in a Willshaw net

A neural network of Willshaw type is investigated under the crucial assumption of no overlap betweenp stored patterns covering the net completely. Taking the thermodynamic limit withp remaining finite a first order phase transition is found, indicating the transition from population of one stored pattern to equal population of all stored patterns. The approximation used to keep mathematics tractable is of Gaussian type.     

The correlation functions near intermittency in a one-dimensional Piecewise parabolic map

Piecewise parabolic maps constitute a family of maps in the fully developed chaotic state and depending on a parameter that can be smoothly tuned to a weakly intermittent situation. Approximate analytic expressions are derived for the corresponding correlation functions. These expressions produce power-law decay at intermittency and a crossover from power-law decay to exponential decay below intermittency. It is shown that the scaling functions and the exponent of the power law depend on the kind of the correlations.     

More relaxed condition for dynamics of discrete time delayed Hopfield neural networks

The dynamics of discrete time delayed Hopfield neural networks is investigated. By using a difference inequality combining with the linear matrix inequality, a sufficient condition ensuring global exponential stability of the unique equilibrium point of the networks is found. The result obtained holds not only for constant delay but also for time-varying delays.     

More relaxed condition for dynamics of discrete time delayed Hopfield neural networks

The dynamics of discrete time delayed Hopfield neural networks is investigated. By using a difference inequality combining with the linear matrix inequality, a sufficient condition ensuring global exponential stability of the unique equilibrium point of the networks is found. The result obtained holds not only for constant delay but also for time-varying delays.     

Neural Network Used for the Fusion of Predictions Obtained by the K-Nearest Neighbors Algorithm Based on Independent Data Sources

The article concerns the problem of classification based on independent data sets—local decision tables. The aim of the paper is to propose a classification model for dispersed data using a modified k-nearest neighbors algorithm and a neural network. A neural network, more specifically a multilayer perceptron, is used to combine the prediction results obtained based on local tables. Prediction results are stored in the measurement level and generated using a modified k-nearest neighbors algorithm. The task of neural networks is to combine these results and provide a common prediction. In the article various structures of neural networks (different number of neurons in the hidden layer) are studied and the results are compared with the results generated by other fusion methods, such as the majority voting, the Borda count method, the sum rule, the method that is based on decision templates and the method that is based on theory of evidence. Based on the obtained results, it was found that the neural network always generates unambiguous decisions, which is a great advantage as most of the other fusion methods generate ties. Moreover, if only unambiguous results were considered, the use of a neural network gives much better results than other fusion methods. If we allow ambiguity, some fusion methods are slightly better, but it is the result of this fact that it is possible to generate few decisions for the test object.     

A Dynamic Evolution Model of Airline Networks

Empirical data show that most of the degree distribution of airline networks assume a double power law. In this work, firstly, we assume cities as sites, Hight between two cities as an edge between two sites, and build a dynamic evolution model for airline networks by improving the BA model, in which the conception of attractiveness plays a decisive role in the course of evolution of the networks. To this end, we discuss whether the attractiveness depends on the site label s or not separately, finally we obtain analytic degree distribution. As a result, if the attractiveness of a site is independent of the degree distribution of sites, which will follow the double power law,otherwise, it will be scale-free. Moreover,degree distribution depends on the parameters of the models, and some parameters are more sensitive than others.     

Invariant Pattern Recognition Using Neural Networks Combined with Optical Wavelet Preprocessor

A novel pattern-recognition system that is invariant against scale-, position- and rotation-changes is proposed. The system is composed of an array of modular neural networks with local space-invariant interconnections (FELSI) [Appl. Opt. 29 (1990) 4790] and a multiwavelet transform preprocessor. The wavelet decomposition of two-dimensional patterns is optically realized by the VanderLugt correlator. To obtain the multiwavelet transforms simultaneously, we synthesize a correlation filter of multiwavelets using computer-generated holograms. The learning process of the FELSI with the techniques of additional noise and weight decay is shown to contribute to the invariant recognition of the system.