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 c...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.展开更多
The demand for adopting neural networks in resource-constrained embedded devices is continuously increasing.Quantization is one of the most promising solutions to reduce computational cost and memory storage on embedd...The demand for adopting neural networks in resource-constrained embedded devices is continuously increasing.Quantization is one of the most promising solutions to reduce computational cost and memory storage on embedded devices.In order to reduce the complexity and overhead of deploying neural networks on Integeronly hardware,most current quantization methods use a symmetric quantization mapping strategy to quantize a floating-point neural network into an integer network.However,although symmetric quantization has the advantage of easier implementation,it is sub-optimal for cases where the range could be skewed and not symmetric.This often comes at the cost of lower accuracy.This paper proposed an activation redistribution-based hybrid asymmetric quantizationmethod for neural networks.The proposedmethod takes data distribution into consideration and can resolve the contradiction between the quantization accuracy and the ease of implementation,balance the trade-off between clipping range and quantization resolution,and thus improve the accuracy of the quantized neural network.The experimental results indicate that the accuracy of the proposed method is 2.02%and 5.52%higher than the traditional symmetric quantization method for classification and detection tasks,respectively.The proposed method paves the way for computationally intensive neural network models to be deployed on devices with limited computing resources.Codes will be available on https://github.com/ycjcy/Hybrid-Asymmetric-Quantization.展开更多
Identifying critical nodes or sets in large-scale networks is a fundamental scientific problem and one of the key research directions in the fields of data mining and network science when implementing network attacks,...Identifying critical nodes or sets in large-scale networks is a fundamental scientific problem and one of the key research directions in the fields of data mining and network science when implementing network attacks, defense, repair and control.Traditional methods usually begin from the centrality, node location or the impact on the largest connected component after node destruction, mainly based on the network structure.However, these algorithms do not consider network state changes.We applied a model that combines a random connectivity matrix and minimal low-dimensional structures to represent network connectivity.By using mean field theory and information entropy to calculate node activity,we calculated the overlap between the random parts and fixed low-dimensional parts to quantify the influence of node impact on network state changes and ranked them by importance.We applied this algorithm and the proposed importance algorithm to the overall analysis and stratified analysis of the C.elegans neural network.We observed a change in the critical entropy of the network state and by utilizing the proposed method we can calculate the nodes that indirectly affect muscle cells through neural layers.展开更多
While progress has been made in information source localization,it has overlooked the prevalent friend and adversarial relationships in social networks.This paper addresses this gap by focusing on source localization ...While progress has been made in information source localization,it has overlooked the prevalent friend and adversarial relationships in social networks.This paper addresses this gap by focusing on source localization in signed network models.Leveraging the topological characteristics of signed networks and transforming the propagation probability into effective distance,we propose an optimization method for observer selection.Additionally,by using the reverse propagation algorithm we present a method for information source localization in signed networks.Extensive experimental results demonstrate that a higher proportion of positive edges within signed networks contributes to more favorable source localization,and the higher the ratio of propagation rates between positive and negative edges,the more accurate the source localization becomes.Interestingly,this aligns with our observation that,in reality,the number of friends tends to be greater than the number of adversaries,and the likelihood of information propagation among friends is often higher than among adversaries.In addition,the source located at the periphery of the network is not easy to identify.Furthermore,our proposed observer selection method based on effective distance achieves higher operational efficiency and exhibits higher accuracy in information source localization,compared with three strategies for observer selection based on the classical full-order neighbor coverage.展开更多
We investigate the impact of pairwise and group interactions on the spread of epidemics through an activity-driven model based on time-dependent networks.The effects of pairwise/group interaction proportion and pairwi...We investigate the impact of pairwise and group interactions on the spread of epidemics through an activity-driven model based on time-dependent networks.The effects of pairwise/group interaction proportion and pairwise/group interaction intensity are explored by extensive simulation and theoretical analysis.It is demonstrated that altering the group interaction proportion can either hinder or enhance the spread of epidemics,depending on the relative social intensity of group and pairwise interactions.As the group interaction proportion decreases,the impact of reducing group social intensity diminishes.The ratio of group and pairwise social intensity can affect the effect of group interaction proportion on the scale of infection.A weak heterogeneous activity distribution can raise the epidemic threshold,and reduce the scale of infection.These results benefit the design of epidemic control strategy.展开更多
Since China’s reform and opening-up,the growing disparity between urban and rural areas and regions has led to massive migration.With China’s Rural Revitalization Strategy and the industrial transfer from the easter...Since China’s reform and opening-up,the growing disparity between urban and rural areas and regions has led to massive migration.With China’s Rural Revitalization Strategy and the industrial transfer from the eastern coastal areas to the inland,the migration direction and pattern of the floating population have undergone certain changes.Using the 2017 China Migrants Dynamic Survey(CMDS),excluding Hong Kong,Macao,and Taiwan regions of China,organized by China’s National Health Commission,the relationship matrix of the floating population is constructed according to the inflow place of the interviewees and their outflow place(the location of the registered residence)in the questionnaire survey.We then apply the complex network model to analyze the migration direction and network pattern of China’s floating population from the city scale.The migration network shows an obvious hierarchical agglomeration.The first-,second-,third-and fourth-tier distribution cities are municipalities directly under the central government,provincial capital cities,major cities in the central and western regions and ordinary cities in all provinces,respectively.The migration trend is from the central and western regions to the eastern coastal areas.The migration network has‘small world’characteristics,forming nine communities.It shows that most node cities in the same community are closely linked and geographically close,indicating that the migration network of floating population is still affected by geographical proximity.Narrowing the urban-rural and regional differences will promote the rational distribution this population.It is necessary to strengthen the reform of the registered residence system,so that the floating population can enjoy urban public services comparable to other populations,and allow migrants to live and work in peace.展开更多
Recommendation Information Systems(RIS)are pivotal in helping users in swiftly locating desired content from the vast amount of information available on the Internet.Graph Convolution Network(GCN)algorithms have been ...Recommendation Information Systems(RIS)are pivotal in helping users in swiftly locating desired content from the vast amount of information available on the Internet.Graph Convolution Network(GCN)algorithms have been employed to implement the RIS efficiently.However,the GCN algorithm faces limitations in terms of performance enhancement owing to the due to the embedding value-vanishing problem that occurs during the learning process.To address this issue,we propose a Weighted Forwarding method using the GCN(WF-GCN)algorithm.The proposed method involves multiplying the embedding results with different weights for each hop layer during graph learning.By applying the WF-GCN algorithm,which adjusts weights for each hop layer before forwarding to the next,nodes with many neighbors achieve higher embedding values.This approach facilitates the learning of more hop layers within the GCN framework.The efficacy of the WF-GCN was demonstrated through its application to various datasets.In the MovieLens dataset,the implementation of WF-GCN in LightGCN resulted in significant performance improvements,with recall and NDCG increasing by up to+163.64%and+132.04%,respectively.Similarly,in the Last.FM dataset,LightGCN using WF-GCN enhanced with WF-GCN showed substantial improvements,with the recall and NDCG metrics rising by up to+174.40%and+169.95%,respectively.Furthermore,the application of WF-GCN to Self-supervised Graph Learning(SGL)and Simple Graph Contrastive Learning(SimGCL)also demonstrated notable enhancements in both recall and NDCG across these datasets.展开更多
Artificial neural networks(ANNs)have led to landmark changes in many fields,but they still differ significantly fromthemechanisms of real biological neural networks and face problems such as high computing costs,exces...Artificial neural networks(ANNs)have led to landmark changes in many fields,but they still differ significantly fromthemechanisms of real biological neural networks and face problems such as high computing costs,excessive computing power,and so on.Spiking neural networks(SNNs)provide a new approach combined with brain-like science to improve the computational energy efficiency,computational architecture,and biological credibility of current deep learning applications.In the early stage of development,its poor performance hindered the application of SNNs in real-world scenarios.In recent years,SNNs have made great progress in computational performance and practicability compared with the earlier research results,and are continuously producing significant results.Although there are already many pieces of literature on SNNs,there is still a lack of comprehensive review on SNNs from the perspective of improving performance and practicality as well as incorporating the latest research results.Starting from this issue,this paper elaborates on SNNs along the complete usage process of SNNs including network construction,data processing,model training,development,and deployment,aiming to provide more comprehensive and practical guidance to promote the development of SNNs.Therefore,the connotation and development status of SNNcomputing is reviewed systematically and comprehensively from four aspects:composition structure,data set,learning algorithm,software/hardware development platform.Then the development characteristics of SNNs in intelligent computing are summarized,the current challenges of SNNs are discussed and the future development directions are also prospected.Our research shows that in the fields of machine learning and intelligent computing,SNNs have comparable network scale and performance to ANNs and the ability to challenge large datasets and a variety of tasks.The advantages of SNNs over ANNs in terms of energy efficiency and spatial-temporal data processing have been more fully exploited.And the development of programming and deployment tools has lowered the threshold for the use of SNNs.SNNs show a broad development prospect for brain-like computing.展开更多
The lethal brain tumor “Glioblastoma” has the propensity to grow over time. To improve patient outcomes, it is essential to classify GBM accurately and promptly in order to provide a focused and individualized treat...The lethal brain tumor “Glioblastoma” has the propensity to grow over time. To improve patient outcomes, it is essential to classify GBM accurately and promptly in order to provide a focused and individualized treatment plan. Despite this, deep learning methods, particularly Convolutional Neural Networks (CNNs), have demonstrated a high level of accuracy in a myriad of medical image analysis applications as a result of recent technical breakthroughs. The overall aim of the research is to investigate how CNNs can be used to classify GBMs using data from medical imaging, to improve prognosis precision and effectiveness. This research study will demonstrate a suggested methodology that makes use of the CNN architecture and is trained using a database of MRI pictures with this tumor. The constructed model will be assessed based on its overall performance. Extensive experiments and comparisons with conventional machine learning techniques and existing classification methods will also be made. It will be crucial to emphasize the possibility of early and accurate prediction in a clinical workflow because it can have a big impact on treatment planning and patient outcomes. The paramount objective is to not only address the classification challenge but also to outline a clear pathway towards enhancing prognosis precision and treatment effectiveness.展开更多
With the advancements of software defined network(SDN)and network function virtualization(NFV),service function chain(SFC)placement becomes a crucial enabler for flexible resource scheduling in low earth orbit(LEO)sat...With the advancements of software defined network(SDN)and network function virtualization(NFV),service function chain(SFC)placement becomes a crucial enabler for flexible resource scheduling in low earth orbit(LEO)satellite networks.While due to the scarcity of bandwidth resources and dynamic topology of LEO satellites,the static SFC placement schemes may cause performance degradation,resource waste and even service failure.In this paper,we consider migration and establish an online migration model,especially considering the dynamic topology.Given the scarcity of bandwidth resources,the model aims to maximize the total number of accepted SFCs while incurring as little bandwidth cost of SFC transmission and migration as possible.Due to its NP-hardness,we propose a heuristic minimized dynamic SFC migration(MDSM)algorithm that only triggers the migration procedure when new SFCs are rejected.Simulation results demonstrate that MDSM achieves a performance close to the upper bound with lower complexity.展开更多
Dear Editor, This letter focuses on the protocol-based non-fragile state estimation problem for a class of recurrent neural networks(RNNs). With the development of communication technology, the networked systems have ...Dear Editor, This letter focuses on the protocol-based non-fragile state estimation problem for a class of recurrent neural networks(RNNs). With the development of communication technology, the networked systems have received particular attentions. The networked system brings advantages such as easy to implement.展开更多
As the key infrastructure of space-ground integrated information networks,satellite communication networks provide high-speed and reliable information transmission.In order to meet the burgeoning service demands of th...As the key infrastructure of space-ground integrated information networks,satellite communication networks provide high-speed and reliable information transmission.In order to meet the burgeoning service demands of the IoT and the Internet,the low-latency LEO satellite network has developed rapidly.However,LEO satellites face inherent problems such as small coverage,fast moving speed and short overhead time,which will be more severe when serving high-dynamic users,e.g.high-speed rails and airplanes.The heterogeneous network composed of GEO,MEO and LEO satellites can provide various services,whose network management and resource allocation are also more challenging.展开更多
Cross-region innovation is widely recognized as an important source of the long-term regional innovation capacity.In the recent past,a growing number of studies has investigated the network structure and mechanisms of...Cross-region innovation is widely recognized as an important source of the long-term regional innovation capacity.In the recent past,a growing number of studies has investigated the network structure and mechanisms of cross-region innovation collaboration in various contexts.However,existing research mainly focuses on physical effects,such as geographical distance and high-speed railway connections.These studies ignore the intangible drivers in a changing environment,the more digitalized economy and the increasingly solidified innovation network structure.Thus,the focus of this study is on estimating determinants of innovation networks,especially on intangible drivers,which have been largely neglected so far.Using city-level data of Chinese patents(excluding Hong Kong,Macao,and Taiwan Province of China),we trace innovation networks across Chinese cities over a long period of time.By integrating a measure on Information and Communications Technology(ICT)development gap and network structural effects into the general proximity framework,this paper explores the changing mechanisms of Chinese innovation networks from a new perspective.The results show that the structure of cross-region innovation networks has changed in China.As mechanisms behind this development,the results confirm the increasingly important role of intangible drivers in Chinese inter-city innovation collaboration when controlling for effects of physical proximity,such as geographical distance.Since digitalization and coordinated development are the mainstream trends in China and other developing countries,these countries'inter-city innovation collaboration patterns will witness dramatic changes under the influence of intangible drivers.展开更多
A significant demand rises for energy-efficient deep neural networks to support power-limited embedding devices with successful deep learning applications in IoT and edge computing fields.An accurate energy prediction...A significant demand rises for energy-efficient deep neural networks to support power-limited embedding devices with successful deep learning applications in IoT and edge computing fields.An accurate energy prediction approach is critical to provide measurement and lead optimization direction.However,the current energy prediction approaches lack accuracy and generalization ability due to the lack of research on the neural network structure and the excessive reliance on customized training dataset.This paper presents a novel energy prediction model,NeurstrucEnergy.NeurstrucEnergy treats neural networks as directed graphs and applies a bi-directional graph neural network training on a randomly generated dataset to extract structural features for energy prediction.NeurstrucEnergy has advantages over linear approaches because the bi-directional graph neural network collects structural features from each layer's parents and children.Experimental results show that NeurstrucEnergy establishes state-of-the-art results with mean absolute percentage error of 2.60%.We also evaluate NeurstrucEnergy in a randomly generated dataset,achieving the mean absolute percentage error of 4.83%over 10 typical convolutional neural networks in recent years and 7 efficient convolutional neural networks created by neural architecture search.Our code is available at https://github.com/NEUSoftGreenAI/NeurstrucEnergy.git.展开更多
Neuromorphic computing,inspired by the human brain,uses memristor devices for complex tasks.Recent studies show that self-organizing random nanowires can implement neuromorphic information processing,enabling data ana...Neuromorphic computing,inspired by the human brain,uses memristor devices for complex tasks.Recent studies show that self-organizing random nanowires can implement neuromorphic information processing,enabling data analysis.This paper presents a model based on these nanowire networks,with an improved conductance variation profile.We suggest using these networks for temporal information processing via a reservoir computing scheme and propose an efficient data encoding method using voltage pulses.The nanowire network layer generates dynamic behaviors for pulse voltages,allowing time series prediction analysis.Our experiment uses a double stochastic nanowire network architecture for processing multiple input signals,outperforming traditional reservoir computing in terms of fewer nodes,enriched dynamics and improved prediction accuracy.Experimental results confirm the high accuracy of this architecture on multiple real-time series datasets,making neuromorphic nanowire networks promising for physical implementation of reservoir computing.展开更多
Two types of one-dimensional(1D)anti-PT-symmetric periodic ring optical waveguide networks,consisting of gain and loss materials,are constructed.The singular optical propagation properties of these networks are invest...Two types of one-dimensional(1D)anti-PT-symmetric periodic ring optical waveguide networks,consisting of gain and loss materials,are constructed.The singular optical propagation properties of these networks are investigated.The results show that the system composed of gain materials exhibits characteristics of ultra-strong transmission and bidirectional reflection.Conversely,the system composed of loss materials demonstrates equal transmittance and reflectance at some frequencies.In both the systems,a new type of total reflection phenomenon is observed.When the imaginary part of the refractive indices of waveguide segments is smaller than 10-5,the system shows bidirectional transparency with the transmittance tending to be 1 and reflectivity to be smaller than 10-8 at some bands.When the refractive indices of the waveguide segments are real,the system will be bidirectional transparent at the full band.These findings may deepen the understanding of anti-PT-symmetric optical systems and optical waveguide networks,and possess potential applications in efficient optical energy storage,ultra-sensitive optical filters,ultra-sensitive all-optical switches,integrated optical chips,stealth physics,and so on.展开更多
The brain is a complex network system in which a large number of neurons are widely connected to each other and transmit signals to each other.The memory characteristic of memristors makes them suitable for simulating...The brain is a complex network system in which a large number of neurons are widely connected to each other and transmit signals to each other.The memory characteristic of memristors makes them suitable for simulating neuronal synapses with plasticity.In this paper,a memristor is used to simulate a synapse,a discrete small-world neuronal network is constructed based on Rulkov neurons and its dynamical behavior is explored.We explore the influence of system parameters on the dynamical behaviors of the discrete small-world network,and the system shows a variety of firing patterns such as spiking firing and triangular burst firing when the neuronal parameterαis changed.The results of a numerical simulation based on Matlab show that the network topology can affect the synchronous firing behavior of the neuronal network,and the higher the reconnection probability and number of the nearest neurons,the more significant the synchronization state of the neurons.In addition,by increasing the coupling strength of memristor synapses,synchronization performance is promoted.The results of this paper can boost research into complex neuronal networks coupled with memristor synapses and further promote the development of neuroscience.展开更多
Urban tourism is considered a complex system,and multiscale exploration of the organizational patterns of attraction networks has become a topical issue in urban tourism,so exploring the multiscale characteristics and...Urban tourism is considered a complex system,and multiscale exploration of the organizational patterns of attraction networks has become a topical issue in urban tourism,so exploring the multiscale characteristics and connection mechanisms of attraction networks is important for understanding the linkages between attractions and even the future destination planning.This paper uses geotagging data to compare the links between attractions in Beijing,China during four different periods:the pre-Olympic period(2004–2007),the Olympic Games and subsequent‘heat period’(2008–2013),the post-Olympic period(2014–2019),and the COVID-19(Corona Virus Disease 2019)pandemic period(2020–2021).The aim is to better understand the evolution and patterns of attraction networks at different scales in Beijing and to provide insights for tourism planning in the destination.The results show that the macro,meso-,and microscales network characteristics of attraction networks have inherent logical relationships that can explain the commonalities and differences in the development process of tourism networks.The macroscale attraction network degree Matthew effect is significant in the four different periods and exhibits a morphological monocentric structure,suggesting that new entrants are more likely to be associated with attractions that already have high value.The mesoscale links attractions according to the common purpose of tourists,and the results of the community segmentation of the attraction networks in the four different periods suggest that the functional polycentric structure describes their clustering effect,and the weak links between clusters result from attractions bound by incomplete information and distance,and the functional polycentric structure with a generally more efficient network of clusters.The pattern structure at the microscale reveals the topological transformation relationship of the regional collaboration pattern,and the attraction network structure in the four different periods has a very similar importance profile structure suggesting that the attraction network has the same construction rules and evolution mechanism,which aids in understanding the attraction network pattern at both macro and micro scales.Important approaches and practical implications for planners and managers are presented.展开更多
Recent advances in deep neural networks have shed new light on physics,engineering,and scientific computing.Reconciling the data-centered viewpoint with physical simulation is one of the research hotspots.The physicsi...Recent advances in deep neural networks have shed new light on physics,engineering,and scientific computing.Reconciling the data-centered viewpoint with physical simulation is one of the research hotspots.The physicsinformedneural network(PINN)is currently the most general framework,which is more popular due to theconvenience of constructing NNs and excellent generalization ability.The automatic differentiation(AD)-basedPINN model is suitable for the homogeneous scientific problem;however,it is unclear how AD can enforce fluxcontinuity across boundaries between cells of different properties where spatial heterogeneity is represented bygrid cells with different physical properties.In this work,we propose a criss-cross physics-informed convolutionalneural network(CC-PINN)learning architecture,aiming to learn the solution of parametric PDEs with spatialheterogeneity of physical properties.To achieve the seamless enforcement of flux continuity and integration ofphysicalmeaning into CNN,a predefined 2D convolutional layer is proposed to accurately express transmissibilitybetween adjacent cells.The efficacy of the proposedmethodwas evaluated through predictions of several petroleumreservoir problems with spatial heterogeneity and compared against state-of-the-art(PINN)through numericalanalysis as a benchmark,which demonstrated the superiority of the proposed method over the PINN.展开更多
Ecosystems generally have the self-adapting ability to resist various external pressures or disturbances,which is always called resilience.However,once the external disturbances exceed the tipping points of the system...Ecosystems generally have the self-adapting ability to resist various external pressures or disturbances,which is always called resilience.However,once the external disturbances exceed the tipping points of the system resilience,the consequences would be catastrophic,and eventually lead the ecosystem to complete collapse.We capture the collapse process of ecosystems represented by plant-pollinator networks with the k-core nested structural method,and find that a sufficiently weak interaction strength or a sufficiently large competition weight can cause the structure of the ecosystem to collapse from its smallest k-core towards its largest k-core.Then we give the tipping points of structure and dynamic collapse of the entire system from the one-dimensional dynamic function of the ecosystem.Our work provides an intuitive and precise description of the dynamic process of ecosystem collapse under multiple interactions,and provides theoretical insights into further avoiding the occurrence of ecosystem collapse.展开更多
基金The authors acknowledge the funding provided by the National Key R&D Program of China(2021YFA1401200)Beijing Outstanding Young Scientist Program(BJJWZYJH01201910007022)+2 种基金National Natural Science Foundation of China(No.U21A20140,No.92050117,No.62005017)programBeijing Municipal Science&Technology Commission,Administrative Commission of Zhongguancun Science Park(No.Z211100004821009)This work was supported by the Synergetic Extreme Condition User Facility(SECUF).
文摘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.
基金The Qian Xuesen Youth Innovation Foundation from China Aerospace Science and Technology Corporation(Grant Number 2022JY51).
文摘The demand for adopting neural networks in resource-constrained embedded devices is continuously increasing.Quantization is one of the most promising solutions to reduce computational cost and memory storage on embedded devices.In order to reduce the complexity and overhead of deploying neural networks on Integeronly hardware,most current quantization methods use a symmetric quantization mapping strategy to quantize a floating-point neural network into an integer network.However,although symmetric quantization has the advantage of easier implementation,it is sub-optimal for cases where the range could be skewed and not symmetric.This often comes at the cost of lower accuracy.This paper proposed an activation redistribution-based hybrid asymmetric quantizationmethod for neural networks.The proposedmethod takes data distribution into consideration and can resolve the contradiction between the quantization accuracy and the ease of implementation,balance the trade-off between clipping range and quantization resolution,and thus improve the accuracy of the quantized neural network.The experimental results indicate that the accuracy of the proposed method is 2.02%and 5.52%higher than the traditional symmetric quantization method for classification and detection tasks,respectively.The proposed method paves the way for computationally intensive neural network models to be deployed on devices with limited computing resources.Codes will be available on https://github.com/ycjcy/Hybrid-Asymmetric-Quantization.
基金Project supported by the National Natural Science Foundation of China (Grant Nos.72071153 and 72231008)Laboratory of Science and Technology on Integrated Logistics Support Foundation (Grant No.6142003190102)the Natural Science Foundation of Shannxi Province (Grant No.2020JM486)。
文摘Identifying critical nodes or sets in large-scale networks is a fundamental scientific problem and one of the key research directions in the fields of data mining and network science when implementing network attacks, defense, repair and control.Traditional methods usually begin from the centrality, node location or the impact on the largest connected component after node destruction, mainly based on the network structure.However, these algorithms do not consider network state changes.We applied a model that combines a random connectivity matrix and minimal low-dimensional structures to represent network connectivity.By using mean field theory and information entropy to calculate node activity,we calculated the overlap between the random parts and fixed low-dimensional parts to quantify the influence of node impact on network state changes and ranked them by importance.We applied this algorithm and the proposed importance algorithm to the overall analysis and stratified analysis of the C.elegans neural network.We observed a change in the critical entropy of the network state and by utilizing the proposed method we can calculate the nodes that indirectly affect muscle cells through neural layers.
基金Project supported by the National Natural Science Foundation of China(Grant Nos.62103375 and 62006106)the Zhejiang Provincial Philosophy and Social Science Planning Project(Grant No.22NDJC009Z)+1 种基金the Education Ministry Humanities and Social Science Foundation of China(Grant Nos.19YJCZH056 and 21YJC630120)the Natural Science Foundation of Zhejiang Province of China(Grant Nos.LY23F030003 and LQ21F020005).
文摘While progress has been made in information source localization,it has overlooked the prevalent friend and adversarial relationships in social networks.This paper addresses this gap by focusing on source localization in signed network models.Leveraging the topological characteristics of signed networks and transforming the propagation probability into effective distance,we propose an optimization method for observer selection.Additionally,by using the reverse propagation algorithm we present a method for information source localization in signed networks.Extensive experimental results demonstrate that a higher proportion of positive edges within signed networks contributes to more favorable source localization,and the higher the ratio of propagation rates between positive and negative edges,the more accurate the source localization becomes.Interestingly,this aligns with our observation that,in reality,the number of friends tends to be greater than the number of adversaries,and the likelihood of information propagation among friends is often higher than among adversaries.In addition,the source located at the periphery of the network is not easy to identify.Furthermore,our proposed observer selection method based on effective distance achieves higher operational efficiency and exhibits higher accuracy in information source localization,compared with three strategies for observer selection based on the classical full-order neighbor coverage.
基金This work was supported by the National Natural Science Foundation of China(Grant No.12072340)the China Postdoctoral Science Foundation(Grant No.2022M720727)the Jiangsu Funding Program for Excellent Postdoctoral Talent(Grant No.2022ZB130).
文摘We investigate the impact of pairwise and group interactions on the spread of epidemics through an activity-driven model based on time-dependent networks.The effects of pairwise/group interaction proportion and pairwise/group interaction intensity are explored by extensive simulation and theoretical analysis.It is demonstrated that altering the group interaction proportion can either hinder or enhance the spread of epidemics,depending on the relative social intensity of group and pairwise interactions.As the group interaction proportion decreases,the impact of reducing group social intensity diminishes.The ratio of group and pairwise social intensity can affect the effect of group interaction proportion on the scale of infection.A weak heterogeneous activity distribution can raise the epidemic threshold,and reduce the scale of infection.These results benefit the design of epidemic control strategy.
基金Under the auspices of the Fund of Social Sciences Research,Ministry of Education of China(No.17YJA840011)。
文摘Since China’s reform and opening-up,the growing disparity between urban and rural areas and regions has led to massive migration.With China’s Rural Revitalization Strategy and the industrial transfer from the eastern coastal areas to the inland,the migration direction and pattern of the floating population have undergone certain changes.Using the 2017 China Migrants Dynamic Survey(CMDS),excluding Hong Kong,Macao,and Taiwan regions of China,organized by China’s National Health Commission,the relationship matrix of the floating population is constructed according to the inflow place of the interviewees and their outflow place(the location of the registered residence)in the questionnaire survey.We then apply the complex network model to analyze the migration direction and network pattern of China’s floating population from the city scale.The migration network shows an obvious hierarchical agglomeration.The first-,second-,third-and fourth-tier distribution cities are municipalities directly under the central government,provincial capital cities,major cities in the central and western regions and ordinary cities in all provinces,respectively.The migration trend is from the central and western regions to the eastern coastal areas.The migration network has‘small world’characteristics,forming nine communities.It shows that most node cities in the same community are closely linked and geographically close,indicating that the migration network of floating population is still affected by geographical proximity.Narrowing the urban-rural and regional differences will promote the rational distribution this population.It is necessary to strengthen the reform of the registered residence system,so that the floating population can enjoy urban public services comparable to other populations,and allow migrants to live and work in peace.
基金This work was supported by the Kyonggi University Research Grant 2022.
文摘Recommendation Information Systems(RIS)are pivotal in helping users in swiftly locating desired content from the vast amount of information available on the Internet.Graph Convolution Network(GCN)algorithms have been employed to implement the RIS efficiently.However,the GCN algorithm faces limitations in terms of performance enhancement owing to the due to the embedding value-vanishing problem that occurs during the learning process.To address this issue,we propose a Weighted Forwarding method using the GCN(WF-GCN)algorithm.The proposed method involves multiplying the embedding results with different weights for each hop layer during graph learning.By applying the WF-GCN algorithm,which adjusts weights for each hop layer before forwarding to the next,nodes with many neighbors achieve higher embedding values.This approach facilitates the learning of more hop layers within the GCN framework.The efficacy of the WF-GCN was demonstrated through its application to various datasets.In the MovieLens dataset,the implementation of WF-GCN in LightGCN resulted in significant performance improvements,with recall and NDCG increasing by up to+163.64%and+132.04%,respectively.Similarly,in the Last.FM dataset,LightGCN using WF-GCN enhanced with WF-GCN showed substantial improvements,with the recall and NDCG metrics rising by up to+174.40%and+169.95%,respectively.Furthermore,the application of WF-GCN to Self-supervised Graph Learning(SGL)and Simple Graph Contrastive Learning(SimGCL)also demonstrated notable enhancements in both recall and NDCG across these datasets.
基金supported by the National Natural Science Foundation of China(Nos.61974164,62074166,62004219,62004220,and 62104256).
文摘Artificial neural networks(ANNs)have led to landmark changes in many fields,but they still differ significantly fromthemechanisms of real biological neural networks and face problems such as high computing costs,excessive computing power,and so on.Spiking neural networks(SNNs)provide a new approach combined with brain-like science to improve the computational energy efficiency,computational architecture,and biological credibility of current deep learning applications.In the early stage of development,its poor performance hindered the application of SNNs in real-world scenarios.In recent years,SNNs have made great progress in computational performance and practicability compared with the earlier research results,and are continuously producing significant results.Although there are already many pieces of literature on SNNs,there is still a lack of comprehensive review on SNNs from the perspective of improving performance and practicality as well as incorporating the latest research results.Starting from this issue,this paper elaborates on SNNs along the complete usage process of SNNs including network construction,data processing,model training,development,and deployment,aiming to provide more comprehensive and practical guidance to promote the development of SNNs.Therefore,the connotation and development status of SNNcomputing is reviewed systematically and comprehensively from four aspects:composition structure,data set,learning algorithm,software/hardware development platform.Then the development characteristics of SNNs in intelligent computing are summarized,the current challenges of SNNs are discussed and the future development directions are also prospected.Our research shows that in the fields of machine learning and intelligent computing,SNNs have comparable network scale and performance to ANNs and the ability to challenge large datasets and a variety of tasks.The advantages of SNNs over ANNs in terms of energy efficiency and spatial-temporal data processing have been more fully exploited.And the development of programming and deployment tools has lowered the threshold for the use of SNNs.SNNs show a broad development prospect for brain-like computing.
文摘The lethal brain tumor “Glioblastoma” has the propensity to grow over time. To improve patient outcomes, it is essential to classify GBM accurately and promptly in order to provide a focused and individualized treatment plan. Despite this, deep learning methods, particularly Convolutional Neural Networks (CNNs), have demonstrated a high level of accuracy in a myriad of medical image analysis applications as a result of recent technical breakthroughs. The overall aim of the research is to investigate how CNNs can be used to classify GBMs using data from medical imaging, to improve prognosis precision and effectiveness. This research study will demonstrate a suggested methodology that makes use of the CNN architecture and is trained using a database of MRI pictures with this tumor. The constructed model will be assessed based on its overall performance. Extensive experiments and comparisons with conventional machine learning techniques and existing classification methods will also be made. It will be crucial to emphasize the possibility of early and accurate prediction in a clinical workflow because it can have a big impact on treatment planning and patient outcomes. The paramount objective is to not only address the classification challenge but also to outline a clear pathway towards enhancing prognosis precision and treatment effectiveness.
基金supported in part by the National Natural Science Foundation of China(NSFC)under grant numbers U22A2007 and 62171010the Open project of Satellite Internet Key Laboratory in 2022(Project 3:Research on Spaceborne Lightweight Core Network and Intelligent Collaboration)the Beijing Natural Science Foundation under grant number L212003.
文摘With the advancements of software defined network(SDN)and network function virtualization(NFV),service function chain(SFC)placement becomes a crucial enabler for flexible resource scheduling in low earth orbit(LEO)satellite networks.While due to the scarcity of bandwidth resources and dynamic topology of LEO satellites,the static SFC placement schemes may cause performance degradation,resource waste and even service failure.In this paper,we consider migration and establish an online migration model,especially considering the dynamic topology.Given the scarcity of bandwidth resources,the model aims to maximize the total number of accepted SFCs while incurring as little bandwidth cost of SFC transmission and migration as possible.Due to its NP-hardness,we propose a heuristic minimized dynamic SFC migration(MDSM)algorithm that only triggers the migration procedure when new SFCs are rejected.Simulation results demonstrate that MDSM achieves a performance close to the upper bound with lower complexity.
基金supported in part by the National Natural Science Foundation of China (U21A2019, 61933007)the Hainan Province Science and Technology Special Fund (ZDYF2022SHFZ105)。
文摘Dear Editor, This letter focuses on the protocol-based non-fragile state estimation problem for a class of recurrent neural networks(RNNs). With the development of communication technology, the networked systems have received particular attentions. The networked system brings advantages such as easy to implement.
文摘As the key infrastructure of space-ground integrated information networks,satellite communication networks provide high-speed and reliable information transmission.In order to meet the burgeoning service demands of the IoT and the Internet,the low-latency LEO satellite network has developed rapidly.However,LEO satellites face inherent problems such as small coverage,fast moving speed and short overhead time,which will be more severe when serving high-dynamic users,e.g.high-speed rails and airplanes.The heterogeneous network composed of GEO,MEO and LEO satellites can provide various services,whose network management and resource allocation are also more challenging.
基金Under the auspices of China Scholarship Council。
文摘Cross-region innovation is widely recognized as an important source of the long-term regional innovation capacity.In the recent past,a growing number of studies has investigated the network structure and mechanisms of cross-region innovation collaboration in various contexts.However,existing research mainly focuses on physical effects,such as geographical distance and high-speed railway connections.These studies ignore the intangible drivers in a changing environment,the more digitalized economy and the increasingly solidified innovation network structure.Thus,the focus of this study is on estimating determinants of innovation networks,especially on intangible drivers,which have been largely neglected so far.Using city-level data of Chinese patents(excluding Hong Kong,Macao,and Taiwan Province of China),we trace innovation networks across Chinese cities over a long period of time.By integrating a measure on Information and Communications Technology(ICT)development gap and network structural effects into the general proximity framework,this paper explores the changing mechanisms of Chinese innovation networks from a new perspective.The results show that the structure of cross-region innovation networks has changed in China.As mechanisms behind this development,the results confirm the increasingly important role of intangible drivers in Chinese inter-city innovation collaboration when controlling for effects of physical proximity,such as geographical distance.Since digitalization and coordinated development are the mainstream trends in China and other developing countries,these countries'inter-city innovation collaboration patterns will witness dramatic changes under the influence of intangible drivers.
基金supported by the Natural Science Foundation of Liaoning Province(2020-BS-054)the Fundamental Research Funds for the Central Universities(N2017005)the National Natural Science Foundation of China(62162050).
文摘A significant demand rises for energy-efficient deep neural networks to support power-limited embedding devices with successful deep learning applications in IoT and edge computing fields.An accurate energy prediction approach is critical to provide measurement and lead optimization direction.However,the current energy prediction approaches lack accuracy and generalization ability due to the lack of research on the neural network structure and the excessive reliance on customized training dataset.This paper presents a novel energy prediction model,NeurstrucEnergy.NeurstrucEnergy treats neural networks as directed graphs and applies a bi-directional graph neural network training on a randomly generated dataset to extract structural features for energy prediction.NeurstrucEnergy has advantages over linear approaches because the bi-directional graph neural network collects structural features from each layer's parents and children.Experimental results show that NeurstrucEnergy establishes state-of-the-art results with mean absolute percentage error of 2.60%.We also evaluate NeurstrucEnergy in a randomly generated dataset,achieving the mean absolute percentage error of 4.83%over 10 typical convolutional neural networks in recent years and 7 efficient convolutional neural networks created by neural architecture search.Our code is available at https://github.com/NEUSoftGreenAI/NeurstrucEnergy.git.
基金Project supported by the National Natural Science Foundation of China (Grant Nos. U20A20227,62076208, and 62076207)Chongqing Talent Plan “Contract System” Project (Grant No. CQYC20210302257)+3 种基金National Key Laboratory of Smart Vehicle Safety Technology Open Fund Project (Grant No. IVSTSKL-202309)the Chongqing Technology Innovation and Application Development Special Major Project (Grant No. CSTB2023TIAD-STX0020)College of Artificial Intelligence, Southwest UniversityState Key Laboratory of Intelligent Vehicle Safety Technology
文摘Neuromorphic computing,inspired by the human brain,uses memristor devices for complex tasks.Recent studies show that self-organizing random nanowires can implement neuromorphic information processing,enabling data analysis.This paper presents a model based on these nanowire networks,with an improved conductance variation profile.We suggest using these networks for temporal information processing via a reservoir computing scheme and propose an efficient data encoding method using voltage pulses.The nanowire network layer generates dynamic behaviors for pulse voltages,allowing time series prediction analysis.Our experiment uses a double stochastic nanowire network architecture for processing multiple input signals,outperforming traditional reservoir computing in terms of fewer nodes,enriched dynamics and improved prediction accuracy.Experimental results confirm the high accuracy of this architecture on multiple real-time series datasets,making neuromorphic nanowire networks promising for physical implementation of reservoir computing.
基金Project supported by the National Natural Science Foundation of China(Grant Nos.11674107,61475049,11775083,61774062,and 61771205).
文摘Two types of one-dimensional(1D)anti-PT-symmetric periodic ring optical waveguide networks,consisting of gain and loss materials,are constructed.The singular optical propagation properties of these networks are investigated.The results show that the system composed of gain materials exhibits characteristics of ultra-strong transmission and bidirectional reflection.Conversely,the system composed of loss materials demonstrates equal transmittance and reflectance at some frequencies.In both the systems,a new type of total reflection phenomenon is observed.When the imaginary part of the refractive indices of waveguide segments is smaller than 10-5,the system shows bidirectional transparency with the transmittance tending to be 1 and reflectivity to be smaller than 10-8 at some bands.When the refractive indices of the waveguide segments are real,the system will be bidirectional transparent at the full band.These findings may deepen the understanding of anti-PT-symmetric optical systems and optical waveguide networks,and possess potential applications in efficient optical energy storage,ultra-sensitive optical filters,ultra-sensitive all-optical switches,integrated optical chips,stealth physics,and so on.
基金Project supported by the Key Projects of Hunan Provincial Department of Education (Grant No.23A0133)the Natural Science Foundation of Hunan Province (Grant No.2022JJ30572)the National Natural Science Foundations of China (Grant No.62171401)。
文摘The brain is a complex network system in which a large number of neurons are widely connected to each other and transmit signals to each other.The memory characteristic of memristors makes them suitable for simulating neuronal synapses with plasticity.In this paper,a memristor is used to simulate a synapse,a discrete small-world neuronal network is constructed based on Rulkov neurons and its dynamical behavior is explored.We explore the influence of system parameters on the dynamical behaviors of the discrete small-world network,and the system shows a variety of firing patterns such as spiking firing and triangular burst firing when the neuronal parameterαis changed.The results of a numerical simulation based on Matlab show that the network topology can affect the synchronous firing behavior of the neuronal network,and the higher the reconnection probability and number of the nearest neurons,the more significant the synchronization state of the neurons.In addition,by increasing the coupling strength of memristor synapses,synchronization performance is promoted.The results of this paper can boost research into complex neuronal networks coupled with memristor synapses and further promote the development of neuroscience.
基金Under the auspices of the National Natural Science Foundation of China(No.41971202)the National Natural Science Foundation of China(No.42201181)the Fundamental research funding targets for central universities(No.2412022QD002)。
文摘Urban tourism is considered a complex system,and multiscale exploration of the organizational patterns of attraction networks has become a topical issue in urban tourism,so exploring the multiscale characteristics and connection mechanisms of attraction networks is important for understanding the linkages between attractions and even the future destination planning.This paper uses geotagging data to compare the links between attractions in Beijing,China during four different periods:the pre-Olympic period(2004–2007),the Olympic Games and subsequent‘heat period’(2008–2013),the post-Olympic period(2014–2019),and the COVID-19(Corona Virus Disease 2019)pandemic period(2020–2021).The aim is to better understand the evolution and patterns of attraction networks at different scales in Beijing and to provide insights for tourism planning in the destination.The results show that the macro,meso-,and microscales network characteristics of attraction networks have inherent logical relationships that can explain the commonalities and differences in the development process of tourism networks.The macroscale attraction network degree Matthew effect is significant in the four different periods and exhibits a morphological monocentric structure,suggesting that new entrants are more likely to be associated with attractions that already have high value.The mesoscale links attractions according to the common purpose of tourists,and the results of the community segmentation of the attraction networks in the four different periods suggest that the functional polycentric structure describes their clustering effect,and the weak links between clusters result from attractions bound by incomplete information and distance,and the functional polycentric structure with a generally more efficient network of clusters.The pattern structure at the microscale reveals the topological transformation relationship of the regional collaboration pattern,and the attraction network structure in the four different periods has a very similar importance profile structure suggesting that the attraction network has the same construction rules and evolution mechanism,which aids in understanding the attraction network pattern at both macro and micro scales.Important approaches and practical implications for planners and managers are presented.
基金the National Natural Science Foundation of China(No.52274048)Beijing Natural Science Foundation(No.3222037)+1 种基金the CNPC 14th Five-Year Perspective Fundamental Research Project(No.2021DJ2104)the Science Foundation of China University of Petroleum,Beijing(No.2462021YXZZ010).
文摘Recent advances in deep neural networks have shed new light on physics,engineering,and scientific computing.Reconciling the data-centered viewpoint with physical simulation is one of the research hotspots.The physicsinformedneural network(PINN)is currently the most general framework,which is more popular due to theconvenience of constructing NNs and excellent generalization ability.The automatic differentiation(AD)-basedPINN model is suitable for the homogeneous scientific problem;however,it is unclear how AD can enforce fluxcontinuity across boundaries between cells of different properties where spatial heterogeneity is represented bygrid cells with different physical properties.In this work,we propose a criss-cross physics-informed convolutionalneural network(CC-PINN)learning architecture,aiming to learn the solution of parametric PDEs with spatialheterogeneity of physical properties.To achieve the seamless enforcement of flux continuity and integration ofphysicalmeaning into CNN,a predefined 2D convolutional layer is proposed to accurately express transmissibilitybetween adjacent cells.The efficacy of the proposedmethodwas evaluated through predictions of several petroleumreservoir problems with spatial heterogeneity and compared against state-of-the-art(PINN)through numericalanalysis as a benchmark,which demonstrated the superiority of the proposed method over the PINN.
基金Project supported by the National Natural Science Foundation of China(Grant Nos.72071153 and 72231008)the Natural Science Foundation of Shaanxi Province(Grant No.2020JM-486)the Fund of the Key Laboratory of Equipment Integrated Support Technology(Grant No.6142003190102)。
文摘Ecosystems generally have the self-adapting ability to resist various external pressures or disturbances,which is always called resilience.However,once the external disturbances exceed the tipping points of the system resilience,the consequences would be catastrophic,and eventually lead the ecosystem to complete collapse.We capture the collapse process of ecosystems represented by plant-pollinator networks with the k-core nested structural method,and find that a sufficiently weak interaction strength or a sufficiently large competition weight can cause the structure of the ecosystem to collapse from its smallest k-core towards its largest k-core.Then we give the tipping points of structure and dynamic collapse of the entire system from the one-dimensional dynamic function of the ecosystem.Our work provides an intuitive and precise description of the dynamic process of ecosystem collapse under multiple interactions,and provides theoretical insights into further avoiding the occurrence of ecosystem collapse.