下载文档原格式
avc1. IntroductionAVC (Advanced Video Coding) is a video compression standard widely used in various electronic devices and applications. It provides an efficient way to represent and transmit video content by reducing the amount of data required for storage and transmission. This document will provide an overview of AVC, its key features, and its advantages over previous video coding standards.2. Overview of AVCAVC, also known as H.264 or MPEG-4 Part 10, is a video coding standard developed by the ITU-T Video Coding Experts Group (VCEG) and the ISO/IEC Moving Picture Experts Group (MPEG). It was first released in 2003 and quickly became one of the most widely adopted video coding standards.AVC’s primary goal is to provide high-quality video compression while maintaining reasonable compression efficiency. It achieves this by utilizing advanced compression techniques such as motion compensation, transform coding, and entropy coding.3. Key Features of AVC3.1. Motion CompensationMotion compensation is a technique used in video coding to reduce temporal redundancy. It takes advantage of the fact that consecutive frames in a video sequence are often highly correlated. By estimating the motion between frames andtransmitting only the difference data (motion vectors), AVC can achieve significant compression gains.3.2. Transform CodingTransform coding is another essential feature of AVC. It converts video data from the spatial domain to the frequency domain using a mathematical transform called the Discrete Cosine Transform (DCT). By focusing on the most significant frequency components, AVC can efficiently represent the video content with fewer bits.3.3. Entropy CodingEntropy coding is the final step in the AVC encoding process. It utilizes statistical modeling and variable-length coding techniques, such as Huffman coding and arithmetic coding, to assign shorter codes to more frequent video patterns. This further reduces the bit rate required for video transmission.4. Advantages of AVCAVC offers several significant advantages over previous video coding standards:4.1. High Compression EfficiencyAVC provides excellent compression efficiency while maintaining high video quality. It can achieve up to 50% or more compression gain compared to older standards like MPEG-2, without significant loss of visual fidelity.4.2. Broad Device and Application SupportAVC has gained widespread support across various electronic devices and applications. It is compatible with a wide range of platforms and operating systems, making it a versatile and widely adopted video coding standard.4.3. Low Bit Rate RequirementsAVC allows for video streaming over low bit-rate networks, making it suitable for bandwidth-constrained environments. It enables smooth video playback even on limited network connections, such as mobile networks or internet connections with low speed.4.4. Scalability and FlexibilityAVC supports various video resolutions, frame rates, and bit rates, making it highly scalable and flexible. It can adapt to different network conditions and device capabilities, ensuring optimal video quality and playback experience.5. ConclusionAVC is a widely adopted video coding standard renowned for its high compression efficiency and video quality. Its advanced techniques, such as motion compensation, transform coding, and entropy coding, allow for efficient video representation and transmission.With broad device and application support, low bit rate requirements, and scalability, AVC has become the go-to standard for video compression in numerous industries, including broadcasting, streaming, video conferencing, and more. Its versatility and effectiveness continue to drive its popularity and ensure its relevance in the digital video landscape.。
因果知觉促进视听时间整合——来自心理物理法和alpha瞬时频率的证据摘要跨通道感觉整合又称多感觉整合，指个体将不同感觉通道内的信息有效整合为一个统一、连贯、完整的多感觉事件的加工过程。
这一加工过程能够将不同感觉通道输入中枢神经系统的信息整合起来，形成整体性知觉以帮助个体更好地感知和适应环境。
其中，视听整合一直是跨通道感觉整合的关键研究领域，且以往研究发现，时间上邻近的视听刺激更容易被整合。
因此，本研究重点关注视听刺激的时间关系，探讨跨通道感觉整合中的视听时间整合。
而另一方面，个体倾向于将具有因果关系的客体知觉为一个整体性事件，从而以因果知觉的形式影响个体的判断，比如影响客体之间的时间距离知觉。
然而，客体在感知过程中存在的因果关联（或者说因果知觉）是否会影响视听时间整合呢？以往研究并未对此进行深入探讨。
为此，本研究采用两项实验予以检验。
实验一将因果知觉的经典客体碰撞范式和视听整合中的经典同时性判断任务相结合，以视听整合的时间窗作为行为指标，探讨因果知觉是否会影响个体的视听时间整合。
结果发现：在有因果关系的条件下，被试的时间窗要显著窄于无因果关系的条件，即被试对视听刺激的时间间隔知觉敏感性变高，促进了视听时间整合。
这一结果证明，因果知觉确实会促进个体的视听时间整合。
实验二则基于脑电实验，在实验一的基础上探讨因果知觉对视听时间整合产生影响的内部机制。
鉴于以往研究发现神经振荡的alpha波代表着信息处理的时间单位，它与感知觉过程、跨通道感觉整合的时间信息密切相关，并且与因果判断也存在联系。
因此实验二采用与实验一中相同的范式，以神经振荡的瞬时频率作为指标，重点检验alpha瞬时频率是否能够作为神经指标，反映因果知觉影响视听时间整合的神经加工。
该实验结果进一步表明，相比无因果条件，有因果关系的条件下alpha瞬时频率显著增加。
并且，我们对比了theta、beta瞬时频率及ERSP，结果发现三者在有无因果两种条件下均未出现显著差异。
宁波大学学报（理工版）首届中国高校优秀科技期刊奖JOURNAL OF NINGBO UNIVERSITY ( NSEE ) 浙江省优秀科技期刊一等奖一种快速运动矢量场搜索的块匹配运动估计算法摘要: 运动估计作为实时视频编解码中最重要最耗时的部分，大量的研究都是通过减少搜索点数来降低计算量。
而块匹配算法以其简单、高效，便于硬件实现等优点被使用到运动估计中。
针对这一特点，提出一种基于块匹配的快速运动矢量场搜索算法（FMVS）。
FMVS算法通过将视频序列时间相关性与空间相关性相结合，提出的一种新算法。
该算法包括以下五部分：预测搜索起点、动态阈值进行静止块判断、方向性类型判定、运动类型判定及混合模板运用。
对视频标准测试序列的实验结果表明,该算法较MVFAST算法,搜索点数降低30%-50%，对于运动复杂的视频序列峰值信噪比提高0.21dB。
关键词: 运动估计；块匹配算法；运动矢量场；（矢量场自适应搜索）MVFAST；峰值信噪比中图分类号: TP393 文献标识码: A 文章编号:对于视频序列图像，由于相连帧之间存在很大的时间相关性，通过减少时间冗余，可以提高视频编码的效率。
而基于块匹配算法以其简单、高效，便于硬件实现等优点，已经被许多视频编码标准所采纳。
运动估计算法占整个编码器的60%~80%的运算量，很大程度决定编码器的效率。
在块匹配运动估计算法中，全搜索算法精度最高，但是运算量也最大大。
为了解决运算量大，产生了很多快速搜索算法。
一类是快速算法是按照某种搜索策略只对搜索窗口的相关参考点进行计算；如一些经典算法3步法[1]，菱形搜索算法[2]，六边形搜索算法[3]。
菱形搜索算法，六边形搜索算法为了避免局部最优，采用大的搜索模板，但带来了搜索点数的大量增加；而小菱形搜索算法采用小菱形减少搜索点数，但是带来局部最优的问题。
另一类快速搜索算法是利用运动矢量相关性来预测当前运动矢量。
此类算法考虑时域或空域相关预测当前搜索起点，性能优于前一种。
学校代码10699分类号TN919.8密级学号**********题目高效视频编码H.265/HEVC率失真优化关键技术研究作者杨楷芳信息与通信工程学科、专业冯燕指导教师2017年01月申请学位日期西北工业大学博士学位论文（学位研究生）题目：高效视频编码H.265/HEVC 率失真优化关键技术研究作者：杨楷芳学科专业：信息与通信工程指导教师：冯燕2017年01月Title: Research on Rate-distortion Optimization Techniques for H.265/HEVCByYang Kai-fangUnder the Supervision of ProfessorFeng YanA Dissertation Submitted toNorthwestern Polytechnical UniversityIn partial fulfillment of the requirementFor the degree ofDoctor of EngineeringXi’an P. R. ChinaJanuary 2017摘要摘 要近些年，随着互联网技术和多媒体技术的快速发展，高清甚至超高清视频业务逐渐走入人们的生活。
这在引起视频数据量迅速增加的同时也给视频的存储和传输带来了巨大的挑战。
2013年1月，视频编码联合组（Joint Collaborative Team on Video Coding，JCT-VC）发布了新一代的面向高清视频的视频编码标准H.265/HEVC，与其前一代视频编码标准H.264/A VC相比，H.265/HEVC可以在保证同等重建视频主观质量前提下节省50%的码率。
率失真优化是视频编码中提高编码效率的关键技术。
率失真优化基于香农的率失真编码理论，通过权衡编码码率与失真，在满足码率限制的前提下，获得尽量高的重建视频质量，进而提高视频编码的率失真性能。
2020年第12期 信息通信2020 (总第 216 期）INFORMATION&COMMUNICATIONS(Sum.N o 216)结合文本及用户资料数据的微博谣言检测柳先觉，徐义春,董方敏(三峡大学计算机与信息学院，湖北宜昌443002)摘要：社交平台谣言检测问题通常以源帖文本，回复文本为谣言检测的判断依据。
此外，用户相关数据也利于提高谣言 检测准确率。
根据文本内容和回复内容呈现的序列特性,个人资料和微博统计数据多维度的无序性，提出基于自注意力 的卷积神经网络及用户信誉特征谣言检测方法。
该方法利用自注意力和卷积神经网络对源帖以及回复文本进行词级和 句子级别的二级编码获取文本语义特征和谣言事件回帖的时序特征，并通过自注意力和最大池化结合用户个人信息及 微博统计数据编码用户信誉特征进行谣言检测。
在取自微博和推特的两个公开数据集上实验表明：1.结合自注意力的 卷积神经网络序列编码优于单一的卷积神经网络;2.用户信誉特征能有效提高谣言检测结果准确率。
关键词：自注意力机制；卷积神经网络;最大池化;用户资料;谣言检测中图分类号:TP391 文献标识码:A文章编号：1673-1131(2020)12-0039-05Microblogging rumor detection combined with text and user profilesL iu X ianjue,X u Y ichun,Dong Fangm in(College o f Computer and Inform ation，China Three Gorges University，Yichang443002, China) Abstract:The rumor detection problem o f social platforms is usually based on the source post text and reply text.In addition, user-related data also helps improve the accuracy o f rumor detection.Based on the sequence characteristics o f text content and reply content,the m ulti-dimensional disorder o f p ersonal data and m icroblog statistics,this paper pro-posed a self-attention convolutional neural network and user credit feature rumor detection method.The method adopts self-attention and convolutional neural networks to perform word-level and sentence-level coding on source post and reply texts to obtain text semantic features and temporal features o f a rumor er credit features and m icroblog statistics are encoded by self-attention and max pooling through user profiles.Experiments are conducted on two public datasets from Weibo and Twitter,and the results demonstrates that: 1.Convolutional neural network sequence coding combined w ith self-attention is superior to a single convolutional neural network;2.The user credit feature can effectively improve the accuracy o f rumor detection.Key words:self-attention mechanism;convolutional neural network;max pooling;user profiles;rumor detectioni概述根据第43次C N N IC中国互联网发展状况统计报告，截 至2018年12月我国网民规模达到8.29亿，互联网普及率 59.6%,网络平台已经成为信息传播的重要途径'同时，社交 网络中充斥的垃圾信息特别是摇言信息成为日益突出的问题。
i c 王贵振 李建民在进行学习、记忆、思维及问题解决等高级认知活动时，人 们需要一个暂时的信息加工与存储机制，它能够保存被激活的信息表征，以备进一步加工之用，1974年 B a d d e l e y 等［1］称这种 机制为工作记忆（w o r k i n g m e m o r y ），代表执行工作记忆的典型 反应为延迟反应。
在功能上，工作记忆包括执行控制和贮存激 活表征两个方面。
表征是暂时或永久储存在神经元网络里的 符号码。
从猴演示的延迟反应任务中得到的两个关键发现说明了前额叶（P F C ）在工作记忆中的重要作用，其中尤其是背侧 前额 叶（d o r s l a t e r a l p r e f r o n t a l c o r t e x ，D L P F C ）作 用。
首 先， D L P F C 主要沟回的损伤引起了延迟依赖的损害［2!4］。
其次，在 延迟反应任务的记忆间期，在 D L P F C 经常可以观察到持久固 定水平的神经冲动发放［5!7］。
本文主要通过以下三个方面来论述 D L P F C 在工作记忆中的作用进展。
1 D L P F C 功能模型明确的证据说明，延迟期间的活化事实上和储存相关，而不是 保持其他的一些相关过程。
如果延迟期间的活化反映了储存的表征，那么一个可能期 望就是记忆全过程的活化应该一直持续到它能指示一个反应。
在一些猴的 P F C 单元记录［5!7，13，18］事件相关f M R I 研究中，通过延长工作记忆任务中的记忆间期，已经报道 D L P F C 活性确实扩展到整个延迟阶段［19!22］。
这些结果和假设 D L P F C 在足够长的时间 内 储 存 活 化 的 表 征 来 指 导 合 适 的 行 为 是 一 致 的。
当然也可以这样解释，就是持续活化反映了把注意投向存在别 处的相关表征的进程。
另一个可能期望就是增加记忆负荷，延 迟期间的活化也应该增加。
I G I T C W技术 研究Technology Study14DIGITCW2023.10随着物联网、5G 、AI 等技术的飞速发展，数据产生的速度和数量都在爆炸式增长[1]，这大大增加了对高效、低时延的通信传输技术的需求。
边缘计算作为一种新型的计算范式，因其能够在靠近数据源的地方完成数据处理，从而大大减少了延迟，提高了数据处理的效率，得到了广泛的关注和研究[2]。
边缘计算不仅能够处理离散的、由边缘设备产生的大量数据，还能够快速响应服务请求，满足实时性的需求[3]。
尤其在一些对时延敏感的应用中，如自动驾驶、远程医疗、智能制造等，边缘计算展现出了无可比拟的优势。
然而，尽管边缘计算具有显著的优势，如何将其与通信技术相结合，实现高效、低时延的数据传输，仍然是一个重要而且具有挑战性的问题。
因此，本文将重点研究基于边缘计算的高效低时延通信传输技术，详细介绍边缘计算和通信技术的总框架，探讨结合方式，以及如何通过优化技术策略实现高效、低时延的数据传输。
希望本文的研究能为边缘计算和通信技术的进一步发展提供一些有价值的思考和参考。
1 基于边缘计算的传输架构基于边缘计算的传输架构由网络服务、核心网EPC 、移动中继节点、汇聚节点以及MEC 服务器（多接入边缘计算）组成。
如图1所示。
网络服务负责管理和控制边缘网络，包括边缘服务器、边缘操作系统、边缘应用程序、边缘云平台和传输协议栈[4]。
EPC 是边缘网络中的一个关键组成部分，它负责管理和配置边缘网络，并提供网络配置、性能监测、安全管理等功能。
移动中继节点负责在移动设备和汇聚节点之间传递数据，并支持多跳、协作传基于边缘计算的高效低时延通信传输技术研究郑 毅（北京华麒通信科技有限公司，北京 100080）摘要：近年来，边缘计算凭借其在灵活性、高效性和可靠性方面的优势，已经成为5G通信研究的热点之一。
边缘计算与5G通信相结合可以进一步提高通信传输的效率和质量。
文章提出了基于边缘计算的高效低时延通信传输技术，对基于边缘计算的高效低时延通信传输技术进行了深入研究，探讨边缘计算与5G通信相结合时的特点和优势，构架了总体传输框架，发现了该技术在实际应用中的问题和挑战，希望能够实现高效低时延的网络通信传输技术。
不良反应信号挖掘流程Adverse drug reactions (ADRs) are a significant concern in healthcare as they can lead to patient harm, increased healthcare costs, and regulatory issues. Therefore, it is crucial to have an effective process for detecting and monitoring ADR signals. In this response, I will discuss the problem of ADR signal mining and outline the requirements for an efficient ADR signal mining workflow.ADRs can arise from various sources, including clinical trials, post-marketing surveillance, and spontaneous reporting systems. The first step in the ADR signal mining process is data collection. This involves gathering information from diverse sources, such as electronic health records, patient reports, and social media platforms. The collected data should be comprehensive and cover a wide range of patient populations and drug exposures.Once the data is collected, the next step is data preprocessing. This involves cleaning the data, removingduplicates, and standardizing the format. It is important to ensure data quality and integrity to minimize false signals and improve the accuracy of the analysis. Additionally, data preprocessing may involve coding the reported adverse events using standardized medical terminology, such as the Medical Dictionary for Regulatory Activities (MedDRA).After data preprocessing, the data is ready for analysis. Various statistical and data mining techniques can be employed to identify potential ADR signals. One commonly used approach is disproportionality analysis, which compares the observed number of ADR reports for a specific drug-event combination with the expected number based on background rates. Other methods include time-to-onset analysis, which examines the temporal relationship between drug exposure and the onset of adverse events, and signal detection algorithms, such as the Bayesian Confidence Propagation Neural Network (BCPNN).Once potential ADR signals are identified, they need to be further evaluated and validated. This involvesconducting detailed clinical assessments, reviewing relevant literature, and consulting domain experts. The goal is to determine the causality and clinicalsignificance of the identified signals. Validated signals are then reported to regulatory authorities and healthcare professionals for appropriate action, such as updating drug labels, issuing safety alerts, or implementing risk management strategies.Finally, continuous monitoring and surveillance are essential to ensure the timely detection of new ADR signals and the assessment of known signals. This involves the establishment of pharmacovigilance systems, which collect and analyze real-world data on drug safety. Feedback mechanisms, such as healthcare professional reporting and patient reporting systems, play a crucial role in this ongoing monitoring process.In conclusion, an effective ADR signal mining workflow involves data collection, preprocessing, analysis, evaluation, and continuous monitoring. It requires a multidisciplinary approach, combining expertise inpharmacology, statistics, data science, and clinical medicine. By implementing such a workflow, healthcare systems can enhance patient safety, improve drug regulation, and ultimately save lives.。
a r X i v :c o n d -m a t /9909379v 1 [c o n d -m a t .d i s -n n ] 27 S e p 1999Fast Response and Temporal Coding on Coherent Oscillations in Small-WorldNetworksLuis go-Fern´a ndez (1),Ram´o n Huerta (1,2),Fernando Corbacho (1),and Juan A.Sig¨u enza (1)(1)Grupo de Neurocomputaci´o n Biol´o gica (GNB),E.T.S.de Ingenier´ıa Inform´a tica,Universidad Aut´o noma de Madrid,28049Madrid (SPAIN).(2)Institute for Nonlinear Science,University of California,San Diego,La Jolla,CA 92093-0402(February 1,2008)We have investigated the role that different connectivity regimes play on the dynamics of a network of Hodgkin-Huxley neurons by computer simulations.The different connectivity topologies exhibit the following features:random connectivity topologies give rise to fast system response yet are unable to produce coherent oscillations in the average activity of the network;on the other hand,regular connectivity topologies give rise to coherent oscillations and temporal coding,but in a temporal scale that is not in accordance with fast signal processing.Finally,small-world (SW)connectivity topologies,which fall between random and regular ones,take advantage of the best features of both,giving rise to fast system response with coherent oscillations along with reproducible temporal coding on clusters of neurons.Our work is the first,to the best of our knowledge,to show the need for a small-world topology in order to obtain all these features in synergy within a biologically plausible time scale.PACS numbers:05.45.-a,87.10.+e,87.18.Bb,87.18.SnIn a recent letter by Watts and Strogatz [1]it was shown that small-world networks enhance signal-propagation speed,computational power,and synchro-nizability.Small-world stands for a network whose con-nectivity topology is placed somewhere between a regular and a completely random connectivity.The main proper-ties of these specific networks are that they can be highly clustered like regular networks and,at the same time,have small path lengths like random ones.Therefore,small-world networks may have properties given neither in regular nor in random networks [2–5].In this letter we have extended Watts and Strogatz’s general framework by introducing dynamical elements in the network nodes.Our source of inspiration is based on a phenomena observed in the olfactory antennal lobe (AL)of the locust discovered by Gilles Laurent and col-laborators [6–9].The AL is a group of around 800neu-rons whose functional role is to relay information from the olfactory receptors to higher areas of the brain for further processing.Three main features have been ob-served in the dynamics of the AL.First,there is a fast response of the AL when the stimulus is presented.Sec-ond,when an odour is presented to the insect,coherent oscillations of 20Hz in the local field potential (LFP)are measured [8].Third,every neuron responds to the odour with some particular timing with respect to the LFP [6].Summarizing:fast response of coherent oscil-lations along with temporal coding are observed.There are also other systems in the brain that present coherent LFP oscillations,hence,hinting to the generality of these phenomena (see [10]for a review).The cooperative behavior of large assemblies of dynam-ical elements has been the subject of many investigations [11–19].In all of them the connectivity between the ele-ments of the network was either regular (local or global all-to-all),or random.However,none of these studies in-corporates a comparative analysis of network dynamics for all the different connectivity topologies.In the present work we pretend to show that in order to provide fast response,coherent oscillations and tem-poral coding a small-world topology is required.We will show that the regular connectivity topology provides a slow response to the external input.Although it is able to produce temporal coding and coherent oscillations,the time of formation of the oscillations would imply much slower responses than those observed in biological sys-tems.On the other hand,for the completely random con-nectivity case the responsiveness of the system is highly increased and temporal variations in clusters activity are present,but the coherent oscillations tipically observed in the LFP are lost.Without these coherent oscillations the AL seems to lose its ability to process the information incoming from the sensors [7].The model we propose for this study is made of an array of non-identical Hodgkin-Huxley elements coupled by excitatory synapses.The unit dynamics is described by the following set of coupled ordinary differential equa-tions:C m ˙Vi =I e (t )−g L ˆV L −g Na m 3h ˆV Na −g K n 4ˆV K +I s (t )(1)˙m =αm (V )(1−m )−βm (V )m (2a)˙h=αh (V )(1−h )−βh (V )h (2b)˙n =αn (V )(1−n )−βn (V )n(2c)where V i represents the membrane potential of unit i;C m is the membrane capacitance per unit area;I e(t) is the external current,which occurs as a pulse of am-plitude I0;I s(t)is the synaptic current;ˆV r=V i−V r, where V r are the equilibrium potentials for the different ionic contributions(r=L,Na,K),and g r are the cor-responding maximum conductances per unit area;h,m, n are the voltage dependent conductances;andα,βare functions of V adjusted to physiological data by voltage clamp techniques.We have used the original functions and parameters employed by Hodgkin and Huxley[20]. The system was integrated using the Runge-Kutta6(5) scheme with variable time step based on[21].The abso-lute error was10−15and the relative error was10−7in all the calculations presented in this letter.The synaptic current I s is given byI s i(t)=g ij r j(t)[V s(t)−E s](3) where i stands for the index of the neuron that receives the synaptic input,j is the neuron from which the synap-tic input is received,and g ij is the maximum conduc-tance,which determines the degree of coupling between the two connected neurons.V s is the postsynaptic po-tential,E s is the synaptic reversal potential and r j(t)is the fraction of bound receptors computed following the method and parameters described by Destexhe et al.[22]. Namely,the dynamics of the bound receptors r is given by the equation:˙r=α[T](1−r)−βr(4) where[T]is the concentration of the transmitter,andα,βare the rise and decay constants,respectively.In this model three different kinds of connectivity pat-terns have been tested:regular,random and small world. To interpolate between regular and random networks we follow the procedure described by Watts and Strogatz[1] which we summarize here for convenience:we start from a ring lattice with N vertices and k edges per vertex, and each edge is rewired at random with probability p. The limits of regularity and randomness are for p=0 and p=1respectively,and the small-world topology lies somewhere in the intermediate region0<p<1. The quantification of the structural properties of these graphs is performed,following Watts and Strogatz[1], using their characteristic path length L(p)and their clus-tering coefficient C(p).L(p)is defined as the number of edges in the shortest path between two vertices,aver-aged over all pairs of vertices.C(p)is defined as follows: suppose that a vertex v has k v neighbours;then at most k v(k v−1)/2edges can exist between them.Let C v de-note the fraction of these allowable edges that actually exist,and define C as the average of C v over all vertices v.Fig.1a replicates that of Watts and Strogatz[1]for ease of reference and to verify our computations.Next we investigate the functional significance of SW topologies for the dynamics of the network.Watts and Strogatz[1]already note that small-world networks of coupled phase oscillators synchronize almost as read-ily as in the mean-field model,despite having orders of magnitude fewer edges.To study the global be-havior of the network we compute its average activityT2−T1T2T1[ V p(t)]2dt(5) wheretiming of action potentials with respect to an ongoing col-lective oscillatory pattern of activity.For instance,when an odour is presented,every neuron in the AL responds to the odourwithsome particulartimingwith respect to the LFP [6].As a measure of this temporal coding,we have divided time in periods of the global average activity,and calculated for each period the quantity:A i (n )=1V (t )]2dt(6)where i represents a particular cluster,n a particular pe-riod of the mean activity[1]D.J.Watts and S.H.Strogatz,Nature 393,440(1998).[2]A.Barrat and M.Weigt,submitted to Eur.Phys.J.B.Also cond-mat/9903411.[3]M.Barth´e l´e my and L.A.N.Amaral,Phys.Rev.Lett.82,3180(1999).[4]M.A.de Menezes,C.F.Moukarzel and T.J.P.Penna,submitted to Phys.Rev.Lett.Also cond-mat/9903426.[5]M.E.J.Newman and D.J.Watts,submitted to Phys.Rev.Lett.Also cond-mat/9903357.[6]M.Wehr and urent,Nature 384,162(1996).[7]urent and H.Davidowitz,Science 265,1872(1994).[8]K.MacLeod and urent,Science 265,976(1996).[9]K.MacLeod,A.B¨a cker and urent,Nature 395,693(1998).[10]C.M.Gray,put.Neurosci.1,11(1994).[11]K.Kaneko,Physica 23D ,436(1986).[12]D.A.Egolf and H.S.Greenside,Nature 369,851(1994).[13]I.S.Aranson,D.Golomb and H.Sompolinsky,Phys.Rev.Lett.68,3495(1992).[14]H.Chat´e ,A.Lemaitre,P.Marq and P.Manneville,Phys-ica 224A ,447(1996).[15]A.V.Gaponov-Grekhov and M.I.Rabinovich,Chaos 6,259(1996).[16]D.Hansel and H.Sompolinsky,Phys.Rev.Lett.68,718(1992).[17]R.Huerta,M.Bazhenov and M.I.Rabinovich,Euro-physics Letters 43(6),719(1998).[18]C.van Vreeswijk and H.Sompolinsky,Science274,1724(1996).[19]C.Fohlmeister,W.Gerstner,R.Ritz and J.L.van Hem-men,Neural Comput.7,1046(1995).[20]A.L.Hodgkin and A.F.Huxley,J.Physiol.117,500(1952).[21]T.E.Hull,W.H.Enright,B.F.Fellen and R.E.Sedg-wick,SIAM J.Num.Anal9,603(1972).[22]A.Destexhe,Z.F.Mainen and T.J.Sejnowski,NeuralComput.6,14(1993).[23]J.J.Hopfield,Nature376,33(1995).[24]G.Deco and B.Schurmann,Phys.Rev.Lett.79,4697(1997).[25]F.Gabbiani,W.Metzner,R.Wessel and C.Koch,Na-ture384,564(1996).[26]N.Brunel,Network:Computation in Neural Systems5,449(1994).FIG.1.(a)Characteristic path length L(p)and clustering coefficient C(p)for the family of randomly rewired graphs, normalized to the values L(0)and C(0)of the regular case.(b)Average activity oscillation amplitudeσ(p)for the whole range of networks,calculated between T1=100and T2=200. Both curves are averages over ten realizations of the simula-tion with parameters N=797,k=30and g=0.015.An input signal I0=1.5was injected,at t=50,to80neurons randomly chosen.FIG.2.Average activity in a network of797neurons.(a)Regular network(p=0.000).(b)Small-world network (p=0.032).(c)Random network(p=1.000).The input onset occurs at t=50and is offset at t=350.All the pa-rameters are as described in Fig.1.FIG.3.Phase diagram which shows the regions of oscilla-tory(clear,highσ)and nonoscillatory(dark,lowσ)activity of the network in the(k,p)plane.The island that appears on the right side indicates that the SW(for some range of val-ues of k)is the only regime capable to produce fast coherent oscillations in the average activity after the presentation of the stimulus.All parameters are as described in the previous figures.FIG.4.(a)-(c)Average activity of three different clusters of neurons promediated over periods of the global mean activ-ity.The simulation corresponds with that of Fig.2b,which lies within the SW region.(d)Average activity of the whole network showing the coherent oscillations over which the ac-tivities of clusters are promediated.FIG.5.Correlation versus noise for two simulations with SW connectivity.(a)p=0.032.(b)p=0.100.The rest of parameters are as described in Fig.1.The plots were ob-tained as follows:we madefive realizations of the simulation for a givenǫ,and calculated the point as a double average;first an average over all possible pairs of realizations for a given cluster,and then an average over clusters.1e−051e−041e−031e−021e−011e+00p12345σ(p)00.20.40.60.81100200300400500time−60−40−200V (t )−60−40−200V (t )−60−40−200V (t )(a)(b)(c)35 30 25 20 15 10212262312362412time−35−30−25−20−15V (t )00.20.40.60.8A 300.20.40.60.8A 200.20.40.60.8A 1(a)(b)(c)(d)1e−051e−041e−031e−021e−011e+00ε0.20.40.60.81C o r r e l a t i on0.00.20.40.60.81.0C o r r e l a t i on。
