基于四叉树的自适应门限分形图象IFS压缩方法

为了克服四叉树分块ＩＦＳ编码过程中,排列块与区域块的匹配误差不考虑输入图象特点这一缺点,提出了一种新的基于迭代函数系统（ＩＦＳ）进行静止灰度图象压缩的方法。该方法是根据当前排列块的方差来确定它与区域块的匹配门限。经分析证明,这种基于自适应门限的ＩＦＳ编码方法与人类视觉系统的特性基本相符,而且实验也证明,这一自适应门限的编码方法提高了ＩＦＳ图象压缩的适应性。     

Syntax-driven Behavior Partitioning for Model-checking of Esterel Programs

We consider the issue of exploiting the structural form of Esterel programs to partition the algorithmic RSS (reachable state space) fix-point construction used in model-checking techniques. The basic idea sounds utterly simple, as seen on the case of sequential composition: in P; Q, first compute entirely the states reached in P, and then only carry on to Q, each time using only the relevant transition relation part. Here a brute-force symbolic breadth-first search would have mixed the exploration of P and Q instead, in case P had different behaviors of various lengths, and that would result in irregular BBD representation of temporary state spaces, a major cause of complexity in symbolic model-checking.Difficulties appear in our decomposition approach when scheduling the different transition parts in presence of parallelism and local signal exchanges. Program blocks (or “Macro-states”) put in parallel can be synchronized in various ways, due to dynamic behaviors, and considering all possibilities may lead to an excessive division complexity. The goal is here to find a satisfactory trade-off between compositional and global approaches. Concretely we use some of the features of the TiGeR BDD library, and heuristic orderings between internal signals, to have the transition relation progress through the program behaviors to get the same effect as a global RSS computation, but with much more localized transition applications. We provide concrete benchmarks showing the usefulness of the approach.     

Real-Time MPEG-2 Video Codec System Using Multiple Digital Signal Processors

A real-time MPEG-2 video codec system, called the KAIST image computing system (KICS), was developed using multiple multimedia video processors (MVP, TMS320C80). The number of computations and optimum implementation of the MPEG-2 algorithm were analyzed for real-time processing. An MVP simulator was used to perform the timing analysis for the real-time MPEG-2 encoding and decoding, and it verified the design of the KICS. Task partitioning and mapping of the MPEG-2 algorithm to multiple MVPs were performed based on load-balancing, minimum data I/O, and the properties of the MVP. The KICS uses five MVPs for MPEG-2 real-time encoding of a main profile at the main level whereas it uses three MVPs for real-time decoding.     

经典Bellman-Ford算法的改进及其实验评估

针对以高效求解有边数限制的最短路问题,对经典Bellman-Ford算法进行了改进．借鉴划分算法的思想,通过减少距离标号的数目,得到了两个改进算法．既然已有的改进算法均不能解决有边数限制的最短路问题,因而本算法是经典Bellman-Ford算法的全新改进．相对于经典Bellman-Ford算法,改进后的算法不仅可有效地节省存储空间, 而且实验表明能显著地提高计算效率．     

结合用户约束文件的高效多FPGA系统分割方法

针对单个现场可编程门阵列(Field Programmable Gate Array,FPGA)可配置逻辑模块(ConfigurableLogic Block,CLB)和输入/输出(I/O)数目受限难以满足大规模复杂电路系统设计需要的问题,提出了一种结合用户约束文件(User Constraint File,UCF)和单个FPGA的EDA(Electronic Design Automation)设计流程,通过部分人为干涉,对多FPGA(multi-FPGA)系统进行分割的方法.应用这种分割方法,可以提高multi-FPGA系统的划分效率,简化设计的复杂度.结果表明:基于这种分割方法的分割工具,结合multi-FPGA系统的硬件结构,可实现一个最多为9芯片multi-FPGA系统的设计和配置.     

A corporate credit rating model using multi-class support vector machines with an ordinal pairwise partitioning approach

Predicting corporate credit-rating using statistical and artificial intelligence (AI) techniques has received considerable research attention in the literature. In recent years, multi-class support vector machines (MSVMs) have become a very appealing machine-learning approach due to their good performance. Until now, researchers have proposed a variety of techniques for adapting support vector machines (SVMs) to multi-class classification, since SVMs were originally devised for binary classification. However, most of them have only focused on classifying samples into nominal categories; thus, the unique characteristic of credit-rating - ordinality - seldom has been considered in the proposed approaches. This study proposes a new type of MSVM classifier (named OMSVM) that is designed to extend the binary SVMs by applying an ordinal pairwise partitioning (OPP) strategy. Our model can efficiently and effectively handle multiple ordinal classes. To validate OMSVM, we applied it to a real-world case of bond rating. We compared the results of our model with those of conventional MSVM approaches and other AI techniques including MDA, MLOGIT, CBR, and ANNs. The results showed that our proposed model improves the performance of classification in comparison to other typical multi-class classification techniques and uses fewer computational resources.     

Network partitioning using harmony search and equivalencing for distributed computing

Power system has a highly interconnected network that requires intense computational effort and resources for centralized control. Distributed computing needs the systems to be partitioned optimally into clusters. The network partitioning is an optimization problem whose objective is to minimize the number of nodes in a cluster and the tie lines between the clusters. Harmony Search(HS) Algorithm is one of the recently developed meta heuristic algorithms that can be applied to optimization problems. In this work, the HS algorithm is applied to the network partitioning problem and power flow based equivalencing is done to represent the external system. Simulation is done on IEEE Standard Test Systems. The algorithm is found to be very effective in partitioning the system hierarchically and the equivalencing method gives accurate results in comparison to the centralized control.     

ENVIRONMENTAL DETERMINANTS OF WATER QUALITY IN BOREAL RIVERS BASED ON PARTITIONING METHODS

Quantitative studies of the relationships between river water quality and environmental variables are needed to improve understanding of the impacts of natural and human factors on aquatic environments. However, multicollinearity between environmental variables can hinder the identification of key factors when water quality–environment relationship is studied using traditional regression methods. This study utilized two alternative statistical methods, variation and hierarchical partitioning, to address these difficulties in studies of river water quality. Using these methods, we explored the effects of catchment physiography, climate and land use variables on total phosphorus and nitrogen, pH, water colour and dissolved oxygen during the years 1995–2005 in 32 boreal rivers in Finland. Catchment physiography and land use explained most of the variation in water quality, especially in phosphorus, nitrogen and water colour data. The strong correlations (r_s > 0.8) between agricultural land use and phosphorus and nitrogen concentrations indicate that water quality is highly affected by agriculture in boreal regions. By determining the environmental drivers of different water quality variables, we can estimate the water quality of different catchments in response to environmental changes and identify areas that are sensitive to global changes. Our study shows that novel statistical methods integrated with geographic information system data and techniques provide deeper insights into water quality–environment relationships than traditional regression, and these should be considered when developing, for example, conservation planning for rivers. Copyright © 2011 John Wiley & Sons, Ltd.     

SoC hardware/software partitioning algorithm for multi-performance index constraints

A new partitioning algorithm based on adaptive chaotic genetic annealing is proposed to solve the hardware/software partitioning problem of System on Chip (SoC) on the multi-performance index. A new objective function with different proportions of punishment based on multi-constraints is presented in this algorithm which is based on the genetic algorithm. The chaos operation is employed to generate the initial population, and the adaptive mutation operator is adopted. Individuals with poor fitness are optimized by the adaptive chaos strategy as the population individuals which tend to be similar. The optimum individuals upon genetic manipulation are optimized by the annealing strategy. Simulation results suggest that the algorithm can reduce the least power consumption by 9.8％ and 4.7％ in the case of 200 nodes, and by 5.6％ and 4％ in the case of 300 nodes, respectively, compared with the simulated annealing algorithm and genetic algorithm.     

指导cache静态划分的程序性能profiling优化技术

对于共享cache的多核处理器,如何管理好各个核对cache的利用,对于充分发挥多核处理器性能是很关键的问题.目前采用的cache替换方法程序间会出现性能干扰,cache静态划分技术则是通过为同时运行的程序分配不同的空间来解决性能干扰问题.为了给程序分配合适大小的cache空间,需要对程序进行性能profiling,即事先多遍运行收集程序在各种cache容量下的性能数据,这种性能profiling方法开销巨大,影响实用.为了解决性能profiling需要多遍运行程序的问题,提出了只需单遍运行的程序性能profiling优化技术.该技术利用在线的phase分析技术识别程序的运行阶段,避免对相同阶段的重复profiling;同时分析程序各phase的性能同cache容量变化的关系趋势,对于性能不敏感的容量变化则不进行profiling,降低开销.在程序运行结束后通过程序各phase在cache各种容量下的性能来估计程序在各容量下的整体性能,以指导cache静态划分.实验表明,该技术的开销仅为7%,而该方法指导的cache划分比未划分时有8%的性能改进,同多遍运行的程序性能profiling指导的cache划分性能相比仅有1%的下降.