Pluggable multitask diffractive neural networks based on cascaded metasurfaces

Optical neural networks have significant advantages in terms of power consumption,parallelism,and high computing speed,which has intrigued extensive attention in both academic and engineering communities.It has been considered as one of the powerful tools in promoting the fields of imaging processing and object recognition.However,the existing optical system architecture cannot be reconstructed to the realization of multi-functional artificial intelligence systems simultaneously.To push the development of this issue,we propose the pluggable diffractive neural networks(P-DNN),a general paradigm resorting to the cascaded metasurfaces,which can be applied to recognize various tasks by switching internal plug-ins.As the proof-of-principle,the recognition functions of six types of handwritten digits and six types of fashions are numerical simulated and experimental demonstrated at near-infrared regimes.Encouragingly,the proposed paradigm not only improves the flexibility of the optical neural networks but paves the new route for achieving high-speed,low-power and versatile artificial intelligence systems.

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.

3D laser scanning strategy based on cascaded deep neural network

A 3D laser scanning strategy based on cascaded deep neural network is proposed for the scanning system converted from 2D Lidar with a pitching motion device. The strategy is aimed at moving target detection and monitoring. Combining the device characteristics, the strategy first proposes a cascaded deep neural network, which inputs 2D point cloud, color image and pitching angle. The outputs are target distance and speed classification. And the cross-entropy loss function of network is modified by using focal loss and uniform distribution to improve the recognition accuracy. Then a pitching range and speed model are proposed to determine pitching motion parameters. Finally, the adaptive scanning is realized by integral separate speed PID. The experimental results show that the accuracies of the improved network target detection box, distance and speed classification are 90.17%, 96.87% and 96.97%, respectively. The average speed error of the improved PID is 0.4239°/s, and the average strategy execution time is 0.1521 s.The range and speed model can effectively reduce the collection of useless information and the deformation of the target point cloud. Conclusively, the experimental of overall scanning strategy show that it can improve target point cloud integrity and density while ensuring the capture of target.

基于改进的Cascade RCNN铸管字符检测算法

由于工业现场采集的铸管字符图像存在背景模糊、字符区域占比小、刻字位置不固定、油漆遮挡等问题,导致现有模型的检测精度难以满足工业现场的需求。针对上述问题,提出改进的Cascade RCNN铸管字符检测算法。首先对特征金字塔进行改进,提出融合小目标增强的特征金字塔(STE-FPN),利用多尺度特征融合的特征增强能力丰富铸管小目标字符的特征信息。其次引入自矫正/池化的ResNeSt(SCP-ResNeSt)作为特征提取网络,利用自矫正卷积和池化操作以提升背景复杂的铸管字符特征提取效率。最后对级联结构进行改进,引进Mask分支结构,可以自适应地检测字符区域并去除干扰区域,优化了检测结果。将改进后的算法在铸管数据集上进行测试,其平均检测精度mAP为99.1%,比原Cascade RCNN算法提高了2.3%,得到的精度表明改进后的性能优于原算法。

Cascades with coupled map lattices in preferential attachment community networks

In this paper, cascading failure is studied by coupled map lattice （CML） methods in preferential attachment community networks. It is found that external perturbation R is increasing with modularity Q growing by simulation. In particular, the large modularity Q can hold off the cascading failure dynamic process in community networks. Furthermore, different attack strategies also greatly affect the cascading failure dynamic process. It is particularly significant to control cascading failure process in real community networks.

Recovery of coupled networks after cascading failures

With society＇s increasing dependence on critical infrastructure such as power grids and communications systems, the robustness of these systems has attracted significant attention.Failure of some nodes can trigger a cascading failure, which completely fragments the network, necessitating recovery efforts to improve robustness of complex systems. Inspired by real-world scenarios, this paper proposes repair models after two kinds of network failures, namely complete and incomplete collapse. In both models, three kinds of repair strategies are possible, including random selection（RS）, node selection based on single network node degree（SD）, and node selection based on double network node degree（DD）. We find that the node correlation in each of the two coupled networks affects repair efficiency. Numerical simulation and analysis results suggest that the repair node ratio and repair strategies may have a significant impact on the economics of the repair process. The results of this study thus provide insight into ways to improve the robustness of coupled networks after cascading failures.

Load-redistribution strategy based on time-varying load against cascading failure of complex network

Cascading failure can cause great damage to complex networks, so it is of great significance to improve the network robustness against cascading failure. Many previous existing works on load-redistribution strategies require global information, which is not suitable for large scale networks, and some strategies based on local information assume that the load of a node is always its initial load before the network is attacked, and the load of the failure node is redistributed to its neighbors according to their initial load or initial residual capacity. This paper proposes a new load-redistribution strategy based on local information considering an ever-changing load. It redistributes the loads of the failure node to its nearest neighbors according to their current residual capacity, which makes full use of the residual capacity of the network. Experiments are conducted on two typical networks and two real networks, and the experimental results show that the new load-redistribution strategy can reduce the size of cascading failure efficiently.

Cascading failures in local-world evolving networks

The local-world (LW) evolving network model shows a transition for the degree distribution between the exponential and power-law distributions, depending on the LW size. Cascading failures under intentional attacks in LW network models with different LW sizes were investigated using the cascading failures load model. We found that the LW size has a significant impact on the network's robustness against deliberate attacks. It is much easier to trigger cascading failures in LW evolving networks with a larger LW size. Therefore, to avoid cascading failures in real networks with local preferential attachment such as the Internet, the World Trade Web and the multi-agent system, the LW size should be as small as possible.

Mitigation Strategy against Cascading Failures on Social Networks

Cascading failures are common phenomena in many of real-world networks,such as power grids,Internet,transportation networks and social networks.It's worth noting that once one or a few users on a social network are unavailable for some reasons,they are more likely to influence a large portion of social network.Therefore,an effective mitigation strategy is very critical for avoiding or reducing the impact of cascading failures.In this paper,we firstly quantify the user loads and construct the processes of cascading dynamics,then elaborate the more reasonable mechanism of sharing the extra user loads with considering the features of social networks,and further propose a novel mitigation strategy on social networks against cascading failures.Based on the realworld social network datasets,we evaluate the effectiveness and efficiency of the novel mitigation strategy.The experimental results show that this mitigation strategy can reduce the impact of cascading failures effectively and maintain the network connectivity better with lower cost.These findings are very useful for rationally advertising and may be helpful for avoiding various disasters of cascading failures on many real-world networks.

Cascading failures in congested complex networks with feedback

In this article, we investigate cascading failures in complex networks by introducing a feedback. To characterize the effect of the feedback, we define a procedure that involves a self-organization of trip distribution during the process of cascading failures. For this purpose, user equilibrium with variable demand is used as an alternative way to determine the traffic flow pattern throughout the network. Under the attack, cost function dynamics are introduced to discuss edge overload in complex networks, where each edge is assigned a finite capacity （controlled by parameter α）. We find that scale-free networks without considering the effect of the feedback are expected to be very sensitive to α as compared with random networks, while this situation is largely improved after introducing the feedback.

Model for cascading failures with adaptive defense in complex networks

This paper investigates cascading failures in networks by considering interplay between the flow dynamic and the network topology, where the fluxes exchanged between a pair of nodes can be adaptively adjusted depending on the changes of the shortest path lengths between them. The simulations on both an artificially created scale-free network and the real network structure of the power grid reveal that the adaptive adjustment of the fluxes can drastically enhance the robustness of complex networks against cascading failures. Particularly, there exists an optimal region where the propagation of the cascade is significantly suppressed and the fluxes supported by the network are maximal. With this understanding, a costless strategy of defense for preventing cascade breakdown is proposed. It is shown to be more effective for suppressing the propagation of the cascade than the recent proposed strategy of defense based on the intentional removal of nodes.

Cascading failure in the wireless sensor scale-free networks

In the practical wireless sensor networks （WSNs）, the cascading failure caused by a failure node has serious impact on the network performance. In this paper, we deeply research the cascading failure of scale-free topology in WSNs. Firstly, a cascading failure model for scale-free topology in WSNs is studied. Through analyzing the influence of the node load on cascading failure, the critical load triggering large-scale cascading failure is obtained. Then based on the critical load, a control method for cascading failure is presented. In addition, the simulation experiments are performed to validate the effectiveness of the control method. The results show that the control method can effectively prevent cascading failure.

Using Genetic Algorithms to Improve the Search of the Weight Space in Cascade－Correlation Neural Network

In this paper, we use the global search characteristics of genetic algorithms to help search the weight space of the neurons in the cascade-correlation architecture. The cascade-correlation learning architecture is a technique of training and building neural networks that starts with a simple network of neurons and adds additional neurons as they are needed to suit a particular problem. In our approach, instead ofmodifying the genetic algorithm to account for convergence problems, we search the weight-space using the genetic algorithm and then apply the gradient technique of Quickprop to optimize the weights. This hybrid algorithm which is a combination of genetic algorithms and cascade-correlation is applied to the two spirals problem. We also use our algorithm in the prediction of the cyclic oxidation resistance of Ni- and Co-base superalloys.

Analysis of overload-based cascading failure in multilayer spatial networks

Many complex networks in real life are embedded in space and most infrastructure networks are interdependent,such as the power system and the transport network.In this paper,we construct two cascading failure models on the multilayer spatial network.In our research,the distance l between nodes within the layer obeys the exponential distribution P(l)~exp(-l/ζ),and the length r of dependency link between layers is defined according to node position.An entropy approach is applied to analyze the spatial network structure and reflect the difference degree between nodes.Two metrics,namely dynamic network size and dynamic network entropy,are proposed to evaluate the spatial network robustness and stability.During the cascading failure process,the spatial network evolution is analyzed,and the numbers of failure nodes caused by different reasons are also counted,respectively.Besides,we discuss the factors affecting network robustness.Simulations demonstrate that the larger the values of average degree<k>,the stronger the network robustness.As the length r decreases,the network performs better.When the probability p is small,asζdecreases,the network robustness becomes more reliable.When p is large,the network robustness manifests better performance asζincreases.These results provide insight into enhancing the robustness,maintaining the stability,and adjusting the difference degree between nodes of the embedded spatiality systems.

Design of Network Cascade Structure for Image Super-Resolution

Image super resolution is an important field of computer research.The current mainstream image super-resolution technology is to use deep learning to mine the deeper features of the image,and then use it for image restoration.However,most of these models mentioned above only trained the images in a specific scale and do not consider the relationships between different scales of images.In order to utilize the information of images at different scales,we design a cascade network structure and cascaded super-resolution convolutional neural networks.This network contains three cascaded FSRCNNs.Due to each sub FSRCNN can process a specific scale image,our network can simultaneously exploit three scale images,and can also use the information of three different scales of images.Experiments on multiple datasets confirmed that the proposed network can achieve better performance for image SR.

Protection Strategies against Cascading Failure for Power Systems of Ring Network

Power grid vulnerability is a key issue with large blackouts, causing power disruption for millions of people. The complexity of power grid, together with excessive number of components, makes it difficult to be modeled. Currently, researchers use complex networks to model and study the performance of power grids. In fact, power grids can be modeled into a complex network by making use of ring network topology, with substations and transmission lines denoted as nodes and edges, respectively. In this paper, three protection schemes are proposed and their effectiveness in protecting the power network under high and low-load attacks is studied. The proposed schemes, namely, Cascaded Load Cut-off (CLC), Cascaded Load Overflow (CLO) and Adaptive-Cascaded Load Overflow (A-CLO), improve the robustness of the power grids, i.e., decrease the value of critical tolerance. Simulation results show that CLC and CLO protection schemes are more effective in improving the robustness of networks than the A-CLO protection scheme. However, the CLC protection scheme is effective only at the expense that certain percentage of the network will have no power supply. Thus, results show that the CLO protection scheme dominates the other protection schemes, CLC and A-CLO, in terms of the robustness of the network, improved with the precise amount of load cut-off determined.

Single-Image Dehazing Based on Two-Stream Convolutional Neural Network

The haze weather environment leads to the deterioration of the visual effect of the image,and it is difficult to carry out the work of the advanced vision task.Therefore,dehazing the haze image is an important step before the execution of the advanced vision task.Traditional dehazing algorithms achieve image dehazing by improving image brightness and contrast or constructing artificial priors such as color attenuation priors and dark channel priors.However,the effect is unstable when dealing with complex scenes.In the method based on convolutional neural network,the image dehazing network of the encoding and decoding structure does not consider the difference before and after the dehazing image,and the image spatial information is lost in the encoding stage.In order to overcome these problems,this paper proposes a novel end-to-end two-stream convolutional neural network for single-image dehazing.The network model is composed of a spatial information feature stream and a highlevel semantic feature stream.The spatial information feature stream retains the detailed information of the dehazing image,and the high-level semantic feature stream extracts the multi-scale structural features of the dehazing image.A spatial information auxiliary module is designed and placed between the feature streams.This module uses the attention mechanism to construct a unified expression of different types of information and realizes the gradual restoration of the clear image with the semantic information auxiliary spatial information in the dehazing network.A parallel residual twicing module is proposed,which performs dehazing on the difference information of features at different stages to improve the model’s ability to discriminate haze images.The peak signal-to-noise ratio(PSNR)and structural similarity are used to quantitatively evaluate the similarity between the dehazing results of each algorithm and the original image.The structure similarity and PSNR of the method in this paper reached 0.852 and 17.557dB on the HazeRD dataset,which were higher than existing comparison algorithms.On the SOTS dataset,the indicators are 0.955 and 27.348dB,which are sub-optimal results.In experiments with real haze images,this method can also achieve excellent visual restoration effects.The experimental results show that the model proposed in this paper can restore desired visual effects without fog images,and it also has good generalization performance in real haze scenes.

Cascading failure in multilayer networks with dynamic dependency groups＊

The cascading failure often occurs in real networks. It is significant to analyze the cascading failure in the complex network research. The dependency relation can change over time. Therefore, in this study, we investigate the cascading fail- ure in multilayer networks with dynamic dependency groups. We construct a model considering the recovery mechanism. In our model, two effects between layers are defined. Under Effect 1, the dependent nodes in other layers will be disabled as long as one node does not belong to the largest connected component in one layer. Under Effect 2, the dependent nodes in other layers will recover when one node belongs to the largest connected component. The theoretical solution of the largest component is deduced and the simulation results verify our theoretical solution. In the simulation, we analyze the influence factors of the network robustness, including the fraction of dependent nodes and the group size, in our model. It shows that increasing the fraction of dependent nodes and the group size will enhance the network robustness under Effect 1. On the contrary, these will reduce the network robustness under Effect 2. Meanwhile, we find that the tightness of the network connection will affect the robustness of networks. Furthermore, setting the average degree of network as 8 is enough to keep the network robust.

Multivariable identification of membrane fouling based on compacted cascade neural network

The membrane fouling phenomenon,reflected with various fouling characterization in the membrane bioreactor(MBR)process,is so complicated to distinguish.This paper proposes a multivariable identification model(MIM)based on a compacted cascade neural network to identify membrane fouling accurately.Firstly,a multivariable model is proposed to calculate multiple indicators of membrane fouling using a cascade neural network,which could avoid the interference of the overlap inputs.Secondly,an unsupervised pretraining algorithm was developed with periodic information of membrane fouling to obtain the compact structure of MIM.Thirdly,a hierarchical learning algorithm was proposed to update the parameters of MIM for improving the identification accuracy online.Finally,the proposed model was tested in real plants to evaluate its efficiency and effectiveness.Experimental results have verified the benefits of the proposed method.

Robustness of community networks against cascading failures with heterogeneous redistribution strategies

Network robustness is one of the core contents of complex network security research.This paper focuses on the robustness of community networks with respect to cascading failures,considering the nodes influence and community heterogeneity.A novel node influence ranking method,community-based Clustering-LeaderRank(CCL)algorithm,is first proposed to identify influential nodes in community networks.Simulation results show that the CCL method can effectively identify the influence of nodes.Based on node influence,a new cascading failure model with heterogeneous redistribution strategy is proposed to describe and analyze node fault propagation in community networks.Analytical and numerical simulation results on cascading failure show that the community attribute has an important influence on the cascading failure process.The network robustness against cascading failures increases when the load is more distributed to neighbors of the same community instead of different communities.When the initial load distribution and the load redistribution strategy based on the node influence are the same,the network shows better robustness against node failure.