目前,主流的基于点击的交互式分割方法对所有用户点击进行无差异的编码.这样的编码方法意味着用户的交互只能给神经网络提供目标的位置信息,且每次点击的影响力是相同的.然而,不同阶段的点击的影响力是不同的.早期的交互用于目标轮廓的...目前,主流的基于点击的交互式分割方法对所有用户点击进行无差异的编码.这样的编码方法意味着用户的交互只能给神经网络提供目标的位置信息,且每次点击的影响力是相同的.然而,不同阶段的点击的影响力是不同的.早期的交互用于目标轮廓的选择,中后期的交互则偏向于对分割结果的局部细节进行微调.因此,应该适当扩大早期点击的影响力,以便更快地获得目标轮廓,同时削弱中后期点击的影响力,以防止因为超调或歧义而影响交互式分割的收敛性.1)本文提出了一种动态盘码(Dynamic Disk Coding,DDC)算法,该算法将用户的每个点击都编码成一个特定半径的圆盘,以此添加关于点击影响力的先验信息;2)本文提出了一个交互式分割网络DDC-Net,通过交互信息预处理模块加强交互信息,并在分割网络的浅层和深层将交互式信息与语义信息进行混合,缓解交互信息随着网络加深而逐渐衰减的问题;3)本文提出了一种改进的模拟训练策略,使得网络在训练时能够充分学习不同编码半径的点击所具备的不同影响力,从而使得提出的方法兼顾收敛速度和收敛性.通过实验表明,本文提出的使用动态盘码的深度交互式分割方法具有科学性和有效性,相较于基线方法,和分别平均取得3.63%和2.44%的提升.展开更多
In recent years,a lot of XOR-based coding schemes have been developed to tolerate double disk failures in Redundant Array of Independent Disks (RAID) architectures,such as EVENODD-code,X-code,B-code and BG-HEDP. Despi...In recent years,a lot of XOR-based coding schemes have been developed to tolerate double disk failures in Redundant Array of Independent Disks (RAID) architectures,such as EVENODD-code,X-code,B-code and BG-HEDP. Despite those researches,the decades-old strategy of Reed-Solomon (RS) code remains the only popular space-optimal Maximum Distance Separable (MDS) code for all but the smallest storage systems. The reason is that all those XOR-based schemes are too difficult to be implemented,it mainly because the coding-circle of those codes vary with the number of disks. By contrast,the coding-circle of RS code is a constant. In order to solve this problem,we develop a new MDS code named Latin code and a cascading scheme based on Latin code. The cascading Latin scheme is a nearly MDS code (with only one or two more parity disks compared with the MDS ones). Nev-ertheless,it keeps the coding-circle of the basic Latin code (i.e. a constant) and the low encod-ing/decoding complexity similar to other parity array codes.展开更多
In distributed cloud storage systems, inevitably there exist multiple node failures at the same time. The existing methods of regenerating codes, including minimum storage regenerating(MSR) codes and minimum bandwidth...In distributed cloud storage systems, inevitably there exist multiple node failures at the same time. The existing methods of regenerating codes, including minimum storage regenerating(MSR) codes and minimum bandwidth regenerating(MBR) codes, are mainly to repair one single or several failed nodes, unable to meet the repair need of distributed cloud storage systems. In this paper, we present locally minimum storage regenerating(LMSR) codes to recover multiple failed nodes at the same time. Specifically, the nodes in distributed cloud storage systems are divided into multiple local groups, and in each local group(4, 2) or(5, 3) MSR codes are constructed. Moreover, the grouping method of storage nodes and the repairing process of failed nodes in local groups are studied. Theoretical analysis shows that LMSR codes can achieve the same storage overhead as MSR codes. Furthermore, we verify by means of simulation that, compared with MSR codes, LMSR codes can reduce the repair bandwidth and disk I/O overhead effectively.展开更多
X-Code is one of the most important redundant array of independent disk （RAID）-6 codes which are capable of tolerating double disk failures. However, the code length of X-Code is restricted to be a prime number, and...X-Code is one of the most important redundant array of independent disk （RAID）-6 codes which are capable of tolerating double disk failures. However, the code length of X-Code is restricted to be a prime number, and such code length restriction of X-Code limits its usage in the real storage systems. Moreover, as a vertical RAID-6 code, X-Code can not be extended easily to an arbitrary code length like horizontal RAID-6 codes. In this paper, a novel and efficient code shortening algorithm for X-Code is proposed to extend X-Code to an arbitrary length. It can be further proved that the code shortening algorithm maintains the maximum-distance-separable （MDS） property of X-Code, and namely, the shortened X-Code is still MDS code with the optimal space efficiency. In the context of the shortening algorithm for X-Code, an in-depth performance analysis on X-Code at consecutive code lengths is conducted, and the impacts of the code shortening algorithm on the performance of X-Code in various performance metrics are revealed.展开更多
It is well known that erasure coding can be used in storage systems to efficiently store data while protecting against failures. Conventionally, the design of erasure codes has focused on the tradeoff between redundan...It is well known that erasure coding can be used in storage systems to efficiently store data while protecting against failures. Conventionally, the design of erasure codes has focused on the tradeoff between redundancy and reliability. Under this criterion, an maximum distance separable（MDS） code has optimal redundancy. In this paper, we address a new class of MDS array codes for tolerating triple node failures by extending the row di- agonal parity（RDP） code, named the RDDP（row double diagonal parity） code. The RDDP code takes advantages of good perform- ances of the RDP code with balanced I/0. A specific triple-erasure decoding algorithm to reduce decoding complexity is depicted by geometric graph, and it is easily implemented by software and hardware. The theoretical analysis shows that the comprehensive properties of the RDDP code are optimal, such as encoding and decoding efficiency, update efficiency and I/0 balance performance.展开更多
文摘目前,主流的基于点击的交互式分割方法对所有用户点击进行无差异的编码.这样的编码方法意味着用户的交互只能给神经网络提供目标的位置信息,且每次点击的影响力是相同的.然而,不同阶段的点击的影响力是不同的.早期的交互用于目标轮廓的选择,中后期的交互则偏向于对分割结果的局部细节进行微调.因此,应该适当扩大早期点击的影响力,以便更快地获得目标轮廓,同时削弱中后期点击的影响力,以防止因为超调或歧义而影响交互式分割的收敛性.1)本文提出了一种动态盘码(Dynamic Disk Coding,DDC)算法,该算法将用户的每个点击都编码成一个特定半径的圆盘,以此添加关于点击影响力的先验信息;2)本文提出了一个交互式分割网络DDC-Net,通过交互信息预处理模块加强交互信息,并在分割网络的浅层和深层将交互式信息与语义信息进行混合,缓解交互信息随着网络加深而逐渐衰减的问题;3)本文提出了一种改进的模拟训练策略,使得网络在训练时能够充分学习不同编码半径的点击所具备的不同影响力,从而使得提出的方法兼顾收敛速度和收敛性.通过实验表明,本文提出的使用动态盘码的深度交互式分割方法具有科学性和有效性,相较于基线方法,和分别平均取得3.63%和2.44%的提升.
基金Supported in part by the National High Technology Re-search and Development Program of China (2008 AA01Z-401)the National Science Foundation of China (No.60903028)+1 种基金Doctoral Fund of Ministry of Education of China (20070055054)Science and Technology De-velopment Plan of Tianjin (08JCYBJC13000)
文摘In recent years,a lot of XOR-based coding schemes have been developed to tolerate double disk failures in Redundant Array of Independent Disks (RAID) architectures,such as EVENODD-code,X-code,B-code and BG-HEDP. Despite those researches,the decades-old strategy of Reed-Solomon (RS) code remains the only popular space-optimal Maximum Distance Separable (MDS) code for all but the smallest storage systems. The reason is that all those XOR-based schemes are too difficult to be implemented,it mainly because the coding-circle of those codes vary with the number of disks. By contrast,the coding-circle of RS code is a constant. In order to solve this problem,we develop a new MDS code named Latin code and a cascading scheme based on Latin code. The cascading Latin scheme is a nearly MDS code (with only one or two more parity disks compared with the MDS ones). Nev-ertheless,it keeps the coding-circle of the basic Latin code (i.e. a constant) and the low encod-ing/decoding complexity similar to other parity array codes.
基金supported in part by the National Natural Science Foundation of China (61640006, 61572188)the Natural Science Foundation of Shaanxi Province, China (2015JM6307, 2016JQ6011)the project of science and technology of Xi’an City (2017088CG/RC051(CADX002))
文摘In distributed cloud storage systems, inevitably there exist multiple node failures at the same time. The existing methods of regenerating codes, including minimum storage regenerating(MSR) codes and minimum bandwidth regenerating(MBR) codes, are mainly to repair one single or several failed nodes, unable to meet the repair need of distributed cloud storage systems. In this paper, we present locally minimum storage regenerating(LMSR) codes to recover multiple failed nodes at the same time. Specifically, the nodes in distributed cloud storage systems are divided into multiple local groups, and in each local group(4, 2) or(5, 3) MSR codes are constructed. Moreover, the grouping method of storage nodes and the repairing process of failed nodes in local groups are studied. Theoretical analysis shows that LMSR codes can achieve the same storage overhead as MSR codes. Furthermore, we verify by means of simulation that, compared with MSR codes, LMSR codes can reduce the repair bandwidth and disk I/O overhead effectively.
基金supported by the National Basic Research Program of China (Grant Nos.2011CB302300, 2011CB302301)the National High-Technology Research and Development Program of China (Grant Nos.2009AA01A401,2009AA01A402)+1 种基金the National Natural Science Foundation of China (Grant Nos.60873028, 60933002, 61025008)the Changjiang Innovation Group of Education of China (Grant No.IRT0725)
文摘X-Code is one of the most important redundant array of independent disk （RAID）-6 codes which are capable of tolerating double disk failures. However, the code length of X-Code is restricted to be a prime number, and such code length restriction of X-Code limits its usage in the real storage systems. Moreover, as a vertical RAID-6 code, X-Code can not be extended easily to an arbitrary code length like horizontal RAID-6 codes. In this paper, a novel and efficient code shortening algorithm for X-Code is proposed to extend X-Code to an arbitrary length. It can be further proved that the code shortening algorithm maintains the maximum-distance-separable （MDS） property of X-Code, and namely, the shortened X-Code is still MDS code with the optimal space efficiency. In the context of the shortening algorithm for X-Code, an in-depth performance analysis on X-Code at consecutive code lengths is conducted, and the impacts of the code shortening algorithm on the performance of X-Code in various performance metrics are revealed.
基金Supported by the National Natural Science Foundation of China(60873216)the Key Project of Sichuan Provincial Department of Education(12ZA223)
文摘It is well known that erasure coding can be used in storage systems to efficiently store data while protecting against failures. Conventionally, the design of erasure codes has focused on the tradeoff between redundancy and reliability. Under this criterion, an maximum distance separable（MDS） code has optimal redundancy. In this paper, we address a new class of MDS array codes for tolerating triple node failures by extending the row di- agonal parity（RDP） code, named the RDDP（row double diagonal parity） code. The RDDP code takes advantages of good perform- ances of the RDP code with balanced I/0. A specific triple-erasure decoding algorithm to reduce decoding complexity is depicted by geometric graph, and it is easily implemented by software and hardware. The theoretical analysis shows that the comprehensive properties of the RDDP code are optimal, such as encoding and decoding efficiency, update efficiency and I/0 balance performance.
