针对现有视频摘要算法以及摘要评价方法未能充分考虑工业智能终端所感知的视频数据特点以及工业智能感知相关应用需求,改写了代表性与多样性两种评价约束,基于此,结合DWConv(Depthwise Convolution)与ConvLSTM(Convolutional Long Short...针对现有视频摘要算法以及摘要评价方法未能充分考虑工业智能终端所感知的视频数据特点以及工业智能感知相关应用需求,改写了代表性与多样性两种评价约束,基于此,结合DWConv(Depthwise Convolution)与ConvLSTM(Convolutional Long Short-Term Memory)设计了一种混合双向多层的工业视频摘要方案。该方案由全局粗粒度特征提取、局部细粒度特征提取、反馈更新以及以查询为驱动的特征融合这4部分组成。为应对工业数据高冗余性、感知的视频噪声大等特点,围绕着ConvLSTM与注意力机制搭建全局特征提取模块;为充分提取视频数据的时空特性,结合注意力机制与DB-DWConvLSTM构建局部特征提取模块;针对工业数据具有的周期性与局部稳定性,借助残差网络思想,设计了融合DWConv反馈模块;为了更加凸显关键帧特征,便于更好的筛选关键帧,研究以查询驱动的特征融合模块。为验证方案的有效性与可行性,将该方案在TVSum与SumMe两个数据集上进行分析验证。实验结果表明:该方案在交叉验证、消融实验以及对比分析中都有着较好的性能。展开更多
With the increasing popularity of cloud computing,privacy has become one of the key problem in cloud security.When data is outsourced to the cloud,for data owners,they need to ensure the security of their privacy;for ...With the increasing popularity of cloud computing,privacy has become one of the key problem in cloud security.When data is outsourced to the cloud,for data owners,they need to ensure the security of their privacy;for cloud service providers,they need some information of the data to provide high QoS services;and for authorized users,they need to access to the true value of data.The existing privacy-preserving methods can't meet all the needs of the three parties at the same time.To address this issue,we propose a retrievable data perturbation method and use it in the privacy-preserving in data outsourcing in cloud computing.Our scheme comes in four steps.Firstly,an improved random generator is proposed to generate an accurate"noise".Next,a perturbation algorithm is introduced to add noise to the original data.By doing this,the privacy information is hidden,but the mean and covariance of data which the service providers may need remain unchanged.Then,a retrieval algorithm is proposed to get the original data back from the perturbed data.Finally,we combine the retrievable perturbation with the access control process to ensure only the authorized users can retrieve the original data.The experiments show that our scheme perturbs date correctly,efficiently,and securely.展开更多
Increased adoption of smartphones leads to the explosive growth of mobile network traffic. Understanding the traffic characteristics of mobile network is important for Intemet service providers （ISPs） to optimize ne...Increased adoption of smartphones leads to the explosive growth of mobile network traffic. Understanding the traffic characteristics of mobile network is important for Intemet service providers （ISPs） to optimize network resources. In this paper, we conduct a detailed measurement study on the hyper text transfer protocol （HTTP） traffic characteristics of cellular network among different operating systems （OSs） as well as different device-types. Firstly, we propose a probability-based method to identify the installed OS of eacb smartphone. Then we analyze the traffic characteristics of these smartphones in terms of OS and device-type based on a dataset across 31 days （a billing cycle）. Finally, we identify the installed apps of each smartphone and compare the usage of apps on the dimensions of OS and device-type. Our measurement study provides insights for network operators to strategize priciag and resource allocation for their cellular data networks.展开更多
User targeting via behavioral analysis is becoming increasingly prevalent in online messaging services.By taking into account users'behavior information such as geographic locations,purchase behaviors,and search h...User targeting via behavioral analysis is becoming increasingly prevalent in online messaging services.By taking into account users'behavior information such as geographic locations,purchase behaviors,and search histories,vendors can deliver messages to users who are more likely to have a strong preference.For example,advertisers can rely on some ad-network for distributing ads to targeted users.However,collecting such personal information for accurate targeting raises severe privacy concerns.In order to incentivize users to participate in such behavioral targeting systems,addressing the privacy concerns becomes of paramount importance.We provide a survey of privacy-preserving user targeting.We present the architectures of user targeting,the security threats faced by user targeting,and existing approaches to privacy-preserving user targeting.Some future research directions are also identified.展开更多
文摘针对现有视频摘要算法以及摘要评价方法未能充分考虑工业智能终端所感知的视频数据特点以及工业智能感知相关应用需求,改写了代表性与多样性两种评价约束,基于此,结合DWConv(Depthwise Convolution)与ConvLSTM(Convolutional Long Short-Term Memory)设计了一种混合双向多层的工业视频摘要方案。该方案由全局粗粒度特征提取、局部细粒度特征提取、反馈更新以及以查询为驱动的特征融合这4部分组成。为应对工业数据高冗余性、感知的视频噪声大等特点,围绕着ConvLSTM与注意力机制搭建全局特征提取模块;为充分提取视频数据的时空特性,结合注意力机制与DB-DWConvLSTM构建局部特征提取模块;针对工业数据具有的周期性与局部稳定性,借助残差网络思想,设计了融合DWConv反馈模块;为了更加凸显关键帧特征,便于更好的筛选关键帧,研究以查询驱动的特征融合模块。为验证方案的有效性与可行性,将该方案在TVSum与SumMe两个数据集上进行分析验证。实验结果表明:该方案在交叉验证、消融实验以及对比分析中都有着较好的性能。
基金supported in part by NSFC under Grant No.61172090National Science and Technology Major Project under Grant 2012ZX03002001+3 种基金Research Fund for the Doctoral Program of Higher Education of China under Grant No.20120201110013Scientific and Technological Project in Shaanxi Province under Grant(No.2012K06-30,No.2014JQ8322)Basic Science Research Fund in Xi'an Jiaotong University(No.XJJ2014049,No.XKJC2014008)Shaanxi Science and Technology Innovation Project(2013SZS16-Z01/P01/K01)
文摘With the increasing popularity of cloud computing,privacy has become one of the key problem in cloud security.When data is outsourced to the cloud,for data owners,they need to ensure the security of their privacy;for cloud service providers,they need some information of the data to provide high QoS services;and for authorized users,they need to access to the true value of data.The existing privacy-preserving methods can't meet all the needs of the three parties at the same time.To address this issue,we propose a retrievable data perturbation method and use it in the privacy-preserving in data outsourcing in cloud computing.Our scheme comes in four steps.Firstly,an improved random generator is proposed to generate an accurate"noise".Next,a perturbation algorithm is introduced to add noise to the original data.By doing this,the privacy information is hidden,but the mean and covariance of data which the service providers may need remain unchanged.Then,a retrieval algorithm is proposed to get the original data back from the perturbed data.Finally,we combine the retrievable perturbation with the access control process to ensure only the authorized users can retrieve the original data.The experiments show that our scheme perturbs date correctly,efficiently,and securely.
文摘Increased adoption of smartphones leads to the explosive growth of mobile network traffic. Understanding the traffic characteristics of mobile network is important for Intemet service providers （ISPs） to optimize network resources. In this paper, we conduct a detailed measurement study on the hyper text transfer protocol （HTTP） traffic characteristics of cellular network among different operating systems （OSs） as well as different device-types. Firstly, we propose a probability-based method to identify the installed OS of eacb smartphone. Then we analyze the traffic characteristics of these smartphones in terms of OS and device-type based on a dataset across 31 days （a billing cycle）. Finally, we identify the installed apps of each smartphone and compare the usage of apps on the dimensions of OS and device-type. Our measurement study provides insights for network operators to strategize priciag and resource allocation for their cellular data networks.
基金This work is supported by the National Natural Science Foundation of China(Nos.61572412,61472316)the Research Grants Council of Hong Kong(Nos.CityU 138513,CityU 11276816)+1 种基金the Innovation and Technology Commission of Hong Kong under ITF Project(No.ITS/307/15)an AWS Education Research Grant,and the Science and Technology Project of Shaanxi Province(Nos.2016ZDJC-05,2013SZS-16).
文摘User targeting via behavioral analysis is becoming increasingly prevalent in online messaging services.By taking into account users'behavior information such as geographic locations,purchase behaviors,and search histories,vendors can deliver messages to users who are more likely to have a strong preference.For example,advertisers can rely on some ad-network for distributing ads to targeted users.However,collecting such personal information for accurate targeting raises severe privacy concerns.In order to incentivize users to participate in such behavioral targeting systems,addressing the privacy concerns becomes of paramount importance.We provide a survey of privacy-preserving user targeting.We present the architectures of user targeting,the security threats faced by user targeting,and existing approaches to privacy-preserving user targeting.Some future research directions are also identified.
