Semiotic Aggregation in Deep Learning

Convolutional neural networks utilize a hierarchy of neural network layers. The statistical aspects of information concentration in successive layers can bring an insight into the feature abstraction process. We analyze the saliency maps of these layers from the perspective of semiotics, also known as the study of signs and sign-using behavior. In computational semiotics, this aggregation operation (known as superization) is accompanied by a decrease of spatial entropy: signs are aggregated into supersign. Using spatial entropy, we compute the information content of the saliency maps and study the superization processes which take place between successive layers of the network. In our experiments, we visualize the superization process and show how the obtained knowledge can be used to explain the neural decision model. In addition, we attempt to optimize the architecture of the neural model employing a semiotic greedy technique. To the extent of our knowledge, this is the first application of computational semiotics in the analysis and interpretation of deep neural networks.     

Optimal convergence in naming game with geography-based negotiation on small-world networks

We propose a negotiation strategy to address the effect of geography on the dynamics of naming games over small-world networks. Communication and negotiation frequencies between two agents are determined by their geographical distance in terms of a parameter characterizing the correlation between interaction strength and the distance. A finding is that there exists an optimal parameter value leading to fastest convergence to global consensus on naming. Numerical computations and a theoretical analysis are provided to substantiate our findings.     

Exploring the origins of the power-law properties of energy landscapes: An egg-box model

Multidimensional potential energy landscapes (PELs) have a Gaussian distribution for the energies of the minima, but at the same time the distribution of the hyperareas for the basins of attraction surrounding the minima follows a power-law. To explore how both these features can simultaneously be true, we introduce an “egg-box” model. In these model landscapes, the Gaussian energy distribution is used as a starting point and we examine whether a power-law basin area distribution can arise as a natural consequence through the swallowing up of higher-energy minima by larger low-energy basins when the variance of this Gaussian is increased sufficiently. Although the basin area distribution is substantially broadened by this process, it is insufficient to generate power-laws, highlighting the role played by the inhomogeneous distribution of basins in configuration space for actual PELs.     

基于声发射和GAN-CNN的铝合金管道法兰 连接松动泄漏检测*

面向管道法兰连接松动引起的泄漏检测需求,为解决数据样本不足和减少特征指标手动选取的繁琐环节。本文,考虑到生成性对抗网络（GAN）作为数据扩充工具,已被证明能够生成与真实数据相似的样本数据。同时,卷积神经网络（CNN）作为一种深度学习方法,为自动提取数据的特征提供了一种有效的方法。开展了基于GAN和CNN的铝合金管道法兰连接松动泄漏检测研究。首先,搭建管道泄漏标定和数据采集实验台,利用声发射技术获取不同等级的原始泄漏信号。其次,采用GAN生成样本数据扩充原始数据。同时,为了评估生成模型的性能,引入统计特评估生成质量。最后,将生成的样本数据与原始数据设置为不同训练集,基于卷积神经网络构建智能分类检测模型,应用于管道泄漏检测。同时,分类检测结果与小样本智能分类方法SVM进行了比较,实验结果表明,基于GAN和CNN构建的智能分类模型可显著提高管道法兰连接松动泄漏检测精度。     

A weight’s agglomerative method for detecting communities in weighted networks based on weight’s similarity

This paper proposes the new definition of the community structure of the weighted networks that groups of nodes in which the edge's weights distribute uniformly but at random between them. It can describe the steady connections between nodes or some similarity between nodes' functions effectively. In order to detect the community structure efficiently, a threshold coefficient κ to evaluate the equivalence of edges' weights and a new weighted modularity based on the weight's similarity are proposed. Then, constructing the weighted matrix and using the agglomerative mechanism, it presents a weight's agglomerative method based on optimizing the modularity to detect communities. For a network with n nodes, the algorithm can detect the community structure in time O(n²log₂ⁿ). Simulations on networks show that the algorithm has higher accuracy and precision than the existing techniques. Furthermore, with the change of κ the algorithm discovers a special hierarchical organization which can describe the various steady connections between nodes in groups.     

Synchronization of complex community networks with nonidentical nodes and adaptive coupling strength

In this Letter, the complex dynamical networks with community structure and nonidentical nodes are considered. The globally asymptotical synchronization of the time-delayed complex community networks onto any uniformly smooth state is studied. Some simple and useful criteria are derived by constructing an effective control scheme and adjusting automatically the adaptive coupling strength. Finally, the developed techniques are applied to two complex community networks which are respectively synchronized to a chaotic trajectory and a periodic orbit, and numerical simulations are provided to show the feasibility of the developed methods.     

Priority-based parameter optimization strategy for reducing the effects of four-wave mixing on WDM system

In order to meet the ultra high speed and ultra long-haul transmission distance in wavelength division multiplexing (WDM) systems, the nonlinear impairment affecting the overall spectral efficiency and system performance should be minimized. This paper proposes a strategy to mitigate the four-wave mixing (FWM) effect in WDM system. The strategy determines the effect of both single and combined effects of second, third, and fourth optimization priority parameters such as fiber length, input power, dispersion, channel spacing, and effective area on FWM power. A comparison study was made under different types of optical fiber such as single-mode fiber (SMF), dispersion shifted fiber, non-zero dispersion fiber, and non-zero dispersion shifted fiber. In addition, the system performance in term of bit-error-rate was calculated in the case of single priority (impact of effective area) and combined priority (impact of effective area, input power, fiber length and channel spacing). The results show that the FWM effect was reduced based on the transmission parameters order of optimization, i.e., priority selection proposed. Moreover, the results indicated that increasing sequentially the effective area, fiber length; channel spacing and decreasing the input power provide the most significant sequence in suppressing the effects of FWM. This priority sequence brought the suppression ratio to approximately 26.3% in SMF, which suppressed the FWM effects up to −50 dBm. In term of BER; the combined priority introduces improvement in BER of 2.31 × 10⁻²⁵ in comparison with single priority that has value of BER 4 × 10⁻¹⁴. Finally, this work suggests that the proposed priority-based parameter optimization strategy is an ideal solution for optimum performance of WDM system.     

A Weighted Evolving Network with Community Size Preferential Attachment

Community structure is an important characteristic in real complex network. It is a network consists of groups of nodes within which links are dense but among which links are sparse. In this paper, the evolving network include node, link and community growth and we apply the community size preferential attachment and strength preferential attachment to a growing weighted network model and utilize weight assigning mechanism from BBV model. The resulting network reflects the intrinsic community structure with generalized power-law distributions of nodes＇ degrees and strengths.     

Optical power equalization for upstream traffic with injection-locked Fabry-Perot lasers in TDM-PON

An optical power equalization of upstream traffic in time-division-multiplexed passive optical network (TDM-PON) based on injection-locked Fabry-Perot lasers has been experimentally investigated. The upstream transmitters with stable spectrum are achieved by using an external injection light source in the optical line terminal (OLT). The different upstream powers can be equalized by injection locking a Fabry-Perot laser diode (FP-LD) biased below threshold current in OLT. The dynamic upstream power range from − 8.5 to − 19.5 db m is reduced to a 1.6 dB maximal power variation, when the uplink signal is directly modulated at 1.25 Gb/s.     

Cascading failure in Watts-Strogatz small-world networks

In this paper, we study the cascading failure in Watts-Strogatz small-world networks. We find that this network model has a heterogeneous betweenness distribution, although its degree distribution is homogeneous. Further study shows that this small-world network is robust to random attack but fragile to intentional attack, in the cascading failure scenario. With comparison to standard random graph and scale-free networks, our result indicates that the robust yet fragile property in the cascading failure scenario is mainly related to heterogeneous betweenness, rather than the network degree distribution. Thus, it suggests that we have to be very careful when we use terms such as homogeneous network and heterogeneous network, unless the distribution we refer to is specified.