摘要：由于基于摩尔定律的器件缩小及其性能增长趋势正在放缓,实现快速和高效节能信息处理的新技术和计算模型越来越被关注。与此同时,越来越多证据表明,对于传统布尔电路和冯诺依曼处理器,超CMOS器件很难与CMOS技术竞争。开发利用新兴器件的独特性能,特别是在非传统电路和架构背景下,具有提供在功率、性能和能力方面数十或百、千倍的改进潜力。为充分发挥超CMOS器件的优势,从器件到电路到体系结架再到算法的跨层设计工作不可或缺。在此背景下,本文研究了嵌入式应用中的高性能神经网络加速器,重点阐述了基于非传统器件技术、电路样式到架构的跨层工作的几种深度神经网络加速器的设计,介绍了应用级基准验证研究工作。讨论表明,跨层设计工作的确可以在实现极大规模高效节能处理方面带来数量级的改进。

摘要：A new motion retargeting algorithm is presented, which adapts the motion capture data to a new character. To make the resulting motion realistic, the physically-based optimization method is adopted. However, the optimization process is difficult to converge to the optimal value because of high complexity of the physical human model. In order to address this problem, an appropriate simplified model automatically determined by a motion analysis technique is utilized, and then motion retargeting with this simplified model as an intermediate agent is implemented. The entire motion retargeting algorithm involves three steps of nonlinearly constrained optimization: forward retargeting, motion scaling and inverse retargeting. Experimental results show the validity of this algorithm.

摘要：Traditional fractal pattern design has some disadvantages such as inability to effectively reflect the characteristics of real scenery and texture. We propose a novel pattern design technique combining fractal geometry and image texture synthesis to solve these problems. We have improved Wei and Levoy (2000)’s texture synthesis algorithm by first using two-dimensional autocorrelation function to analyze the structure and distribution of textures, and then determining the size of L neighborhood. Several special fractal sets were adopted and HSL (Hue, Saturation, and Light) color space was chosen. The fractal structure was used to manipulate the texture synthesis in HSL color space where the pattern’s color can be adjusted conveniently. Experiments showed that patterns with different styles and different color characteristics can be more efficiently generated using the new technique.

摘要：本文提出并实现了一种基于人类语音识别的智能超表面,用于对电磁波束进行可编程调控。该智能超表面平台由数字编码超表面、语音识别模块、单片机和数模转换器(DAC)电路组成,可根据预先存储的语音指令对电磁波进行智能控制。所构建的数字编码超表面包含6×6个超级子单元,每个超级子单元由4×4个嵌入了变容二极管的有源数字单元组成。语音识别模块配合DAC和单片机对语音指令进行识别,并生成对应的电压序列来控制超表面。此外,在超表面的设计过程中引入遗传算法,可有效优化超表面相位分布。为了验证智能超表面平台的性能,演示了雷达散射截面积缩减、涡旋波束生成和波束分裂三种典型功能。所提出的方案为调控电磁波提供了一种新的途径,并在电磁和声学通信之间架起了连接桥梁。

摘要：由于新兴处理器、内存和网络技术的显著进步,百亿亿级系统将在未来几年(2020–2022)推出。随着百亿亿级系统被配置和使用,具有更细粒度、更短时间步长和更大数据量的下一代应用程序将被持续需求。从发展趋势看,2025–2035年间,下一代应用程序将需要超百亿亿级系统。本文关注超百亿亿级系统在网络和通信方面面临的挑战。首先,提出超百亿亿级系统的设想架构。其次,从不同方面阐述面临的挑战,包括多种网络技术、高性能通信和同步协议、加速器和现场可编程门户阵列的集成支持、容错和服务质量支持、能量感知通信方案和协议、软件定义网络以及多种内存和存储器的可扩展通信协议。再次,指出在这些系统上进行支持高性能计算、大数据和深度学习的高效编程模型设计面临的挑战。最后,强调了这些系统的上层共同设计运行时间的关键需求,以实现最优性能和可扩展性。

摘要：XML data can be represented by a tree or graph and the query processing for XML data requires the structural information among nodes. Designing an efficient labeling scheme for the nodes of Order-Sensitive XML trees is one of the important methods to obtain the excellent management of XML data. Previous labeling schemes such as region and prefix often sacrifice updating performance and suffer increasing labeling space when inserting new nodes. To overcome these limitations, in this paper we propose a new labeling idea of separating structure from order. According to the proposed idea, a novel Prime-based Middle Fraction Labeling Scheme(PMFLS) is designed accordingly, in which a series of algorithms are proposed to obtain the structural relationships among nodes and to support updates. PMFLS combines the advantages of both prefix and region schemes in which the structural information and sequential information are separately expressed. PMFLS also supports Order-Sensitive updates without relabeling or recalculation, and its labeling space is stable. Experiments and analysis on several benchmarks are conducted and the results show that PMFLS is efficient in handling updates and also significantly improves the performance of the query processing with good scalability.

摘要：Fuzzy entropy was designed for non convex fuzzy membership function using well known Hamming distance *** proposed fuzzy entropy had the same structure as that of convex fuzzy membership *** procedure of fuzzy entropy was proposed by considering fuzzy membership through distance measure,and the obtained results contained more flexibility than the general fuzzy membership ***,characteristic analyses for non convex function were also *** on the mutual information were carried out through the proposed fuzzy entropy and similarity measure,which was also dual structure of fuzzy *** the illustrative example,mutual information was discussed.

摘要：科学正在进入一个新时代——第五范式——它被认为是知识整合到不同领域的主要特征,是基于无所不在的机器学习系统的计算社区中智能驱动的工作。在此,我们通过在天河一号超级计算机系统上构建的催化材料专门设计的典型平台案例,生动地阐明了第五范式的本质,旨在促进第五范式在其他领域的培养。第五范式平台主要包括模型自动构建(原始数据提取)、指纹自动构建(神经网络特征选择)以及跨学科知识串联的重复迭代(“火山图”)。与分解一起进行的是对迭代中实现的体系结构的性能评估。通过讨论,第五范式的智能驱动平台可以极大地简化和改进研究中极其繁琐和具有挑战性的工作,并通过补偿机器学习中缺少样本和替代一些由于计算资源不足而导致的数值计算来实现数值计算与机器学习之间的相互反馈,从而加速探索过程。在数据驱动的学科中,跨学科专家的协同作用和对动态数据需求的急剧增长仍然是一个挑战。我们相信,对第五范式平台的一瞥可以为其在其他领域的应用铺平道路。

摘要：本文提出了一种利用inverse butterfly网络的自路由特性完成循环移位、短字循环移位等操作的可重构控制信息生成算法。该算法具有高并行度和低成本,此外还支持循环移位、双向循环移位以及短字循环移位操作。据我们所知,这是首个将多种不同类型的循环移位操作统一在一个硬件架构中的算法。本文进一步设计了一种高效可重构循环移位单元,并在SMIC 65-nm工艺下实现了逻辑综合。结果表明:与以往具有相似功能的设计相比,该单元的效能至少提升了23%;当仅支持双向循环移位操作时,该单元具有更小的面积和更低的延迟。

摘要：For a hierarchical cognitive radio network(CRN),the secondary users(SUs) may access the licensed spectrum opportunistically,whenever it is not occupied by the primary users(PUs).An important issue for this kind of CRN is the achievable qualityof-service(QoS) performance,such as traffic transmission delay,which is critical to the SUs' traffic *** this paper,we focus on the delay performance analysis of the SU system and the design of the corresponding optimal access strategy for the case of SUs sharing multiple licensed *** our analysis,the transmission of PU and SU traffic is modeled as M/G/1 *** merging the PU and SU traffic,we propose the model of a priority virtual queue on the licensed *** on this model,we obtain the expected system delay expression for SU traffic through M/G/1 preemptive repeat priority queuing *** the case of multiple licensed channel access,the access strategy is further investigated with respect to the expected system delay for SU *** minimizing the expected transmission delay,the optimal access strategy is modeled as a nonlinear programming problem,which can be resolved by means of the classic Genetic Algorithm(GA).Numerical results validate our analysis and design of an optimal access ***,by considering the time taken by the GA approach,we can also adopt the inverse proportional access strategy to obtain near-optimal results in practice.