分布式求解平行结构类问题中重选区域的研究

本文以平行结构类问题的形式化描述为基础,详细讨论了分布式求解平行结构类问题系统中相邻结点间重迭区域的作用,定义了最小重迭区域的概念,并提出一种新的设立重迭区域方法——“可变法”,最后在分布式运输调度系统上进行了试验研究,结果表明:〈1〉设置重迭区域是必要的;〈2〉最小重迭区域的定义是有意义的;〈3〉与美国麻省大学采用的“固定法”相比,“可变法”保证了求解质量,减少了冗余的计算与通信开销。     

一种双向求解平行结构类问题的方法

平行结构类问题是一类适于分布式求解的人工智能问题．已有的大多数求解方法均采用预期或目标来指导自底向上的问题求解．但这些预期或目标是以局部问题求解状态为基础的，指导性较弱．尽管有的方法（如改进的ＤＶＭＴ结构）允许高层了解，但未给出明确的求解算法．本文提出一种双向求解平行结构类问题的方法，首先根据全局问题求解状态生成预期，指导自底向上的求解，然后根据新产生的假设来验证和修改预期，并用新的预期重新指导求解．该方法不仅提高了预期的指导性，而且使问题求解更为灵活．     

分布式问题求解应用的分类

本文提出一种新的分类法,根据组成结构将分布式问题求解应用背景分为层次结构和平行结构两大类,并同E.H.Durfee等提出的根据求解内容分类法进行了比较,结果表明:根据组成结构分类法反映了问题与求解系统的内在联系,更利于提高研究的通用性。     

基于分布式优化的协同干扰任务分配研究

为保证目标区域干扰覆盖和最小能量消耗的优化目标,建立协同干扰任务分配模型。在分布式协同优化框架下,将集中式任务分配问题,转换为各个虚任务区内小规模的分布式优化问题,采用分解-协调优化模式和启发式遗传算法相结合的方法,实现对各个子区域优化问题的二次迭代求解。仿真结果表明,分布式协同优化方法能够有效降低协同干扰任务分配问题的求解规模,避免“维数灾”,具有可行性。     

求解二维Fredholm积分方程的参数化信赖域方法

给出了求解二维第一类Fredholm积分方程信赖域方法。通过引入正则化参数将离散后的Fredholm积分方程转化带参数的最优化问题,借助于KKT条件将二次信赖域子问题参数化,并进行分析求解,最后给出了数值模拟。     

一种基于分布式编码的卷积优化算法

卷积在统计学、信号处理、图像处理、深度学习等领域有着广泛的应用,且起到了至关重要的作用。在深度神经网络中,使用卷积运算对输入信息进行特征提取的方法是实现神经网络的基础计算单元之一。如何优化卷积的运算速度,提高卷积计算效率一直是亟需探讨的问题。近年来,很多研究指出分布式计算架构可以提高卷积神经网络的计算速度,进而优化深度学习的训练效率,然而由于分布式系统中普遍存在落跑者问题(straggler),该问题可能会拖慢整个系统执行任务的时间,因此该问题也成为了分布式深度学习中一个待解决的问题。文中针对二维卷积计算,结合Winograd算法和分布式编码,提出了一种优化的分布式二维卷积算法。Winograd算法能够有效地加速单次二维卷积计算的速度,分布式编码通过使用一种基于分布式冗余的编码方式能够缓解straggler节点对整个分布式系统计算延迟的影响。因此,提出的分布式二维卷积算法可以在加速二维卷积计算的同时有效缓解分布式系统中的straggler问题,有效提高了分布式卷积的计算效率。     

噪声图像的快速二维Otsu阈值分割

为了提高二维阈值分割法处理速度, 提出了二维Otsu法的快速实现方法。基于二维随机变量的边缘概率分布, 将二维最佳阈值（s*, t*）的求解拆分成两个一维最佳阈值s*和t*的求解; 同时为了改善原算法的分割效果, 引入类内方差的定义, 提出了新的最佳阈值判别式。实验结果表明, 本方法不仅保留了原二维阈值法抗噪性强的特点, 其时间复杂度由O(L⁴)降为O(L), 空间复杂度由S(L²)降为S(L), 且分割错误率低于原二维Otsu法。该方法适合处理高斯噪声图像的快速阈值分割问题。     

基于多层油藏问题负载均衡的并行任务划分S

该文基于分布式并行计算机系统,对一类多层二维二相流油藏数值模拟问题给出了3种任务划分策略-"卷帘"方式、区域分解方式和"卷帘"与区域分解结合的方式,对它们进行了比较,提出了减少求解时间、利于负载均衡和提高并行性能的任务划分方法,并实际应用于有多达72万个网格节点的大规模油藏模拟问题.实算结果表明,该策略划分产生的并行求解任务均衡,有利于加速比的提高.该方法也适用于区域或数据并行的任务划分问题.     

基于多层油藏问题负载均衡的并行任务划分

该文基于分布式并行计算机系统,对一类多层二维二相流油藏数值模拟问题给出了3种任务划分策略—“卷帘”方式、区域分解方式和“卷帘”与区域分解结合的方式,对它们进行了比较,提出了减少求解时间、利于负载均衡和提高并行性能的任务划分方法,并实际应用于有多达72万个网格节点的大规模油藏模拟问题.实算结果表明,该策略划分产生的并行求解任务均衡,有利于加速比的提高.该方法也适用于区域或数据并行的任务划分问题.     

Fredholm积分方程的正则化GMRES算法

利用数值求积公式,对二维第1类Fredholm积分方程进行离散处理,引入正则化GMRES算法,将离散后的积分方程转化为离散适定问题,通过广义极小残余算法得到其数值解。数值模拟结果表明,正则化GMRES算法求解二维第1类Fredholm积分方程计算速度快、精度高。     

用多Agent系统分配具有启动成本的有限资源

提出了基于多Agent系统分配具有启动成本的有限资源的两种方法：集中式求解方法和分布式求解方法。在分布式求解方法中,给出了分布式的连续双向拍卖协议算法,其Agent采用零智慧增强学习策略。实验结果表明：相对于集中式求解方法,使用分布式求解方法是以降低部分效率为代价的,但是当市场的总需求逐渐接近所有卖者能提供的最大资源数时,市场平均效率呈现逐渐递增的趋势。     

Exponential Lower Bounds on the Complexity of?a?Class of?Dynamic Programs for Combinatorial Optimization Problems

We prove exponential lower bounds on the running time of Dynamic Programs (DP) of a certain class for some Combinatorial Optimization Problems. The class of DPs for which we derive the lower bounds is general enough to include well-known DPs for Combinatorial Optimization Problems, such as the ones developed for the Shortest Path Problem, the Knapsack Problem, or the Traveling Salesman Problem. The problems analyzed include the Traveling Salesman Problem (TSP), the Bipartite Matching Problem (BMP), the Min and the Max Cut Problems (MCP), the Min Partition Problem (MPP), and the Min Cost Test Collection Problem (MCTCP). We draw a connection between Dynamic Programs and algorithms for polynomial evaluation. We then derive and use complexity results of polynomial evaluation to prove similar results for Dynamic Programs for the TSP or the BMP. We define a reduction between problems that allows us to generalize these bounds to problems for which either the TSP or the BMP transforms to. Moreover, we show that some standard transformations between problems are of this kind. In this fashion, we extend the lower bounds to other Combinatorial Optimization Problems.     

An efficient distributed algorithm for generating and updating multicast trees

As group applications are becoming widespread, efficient network utilization becomes a growing concern. Multicast transmission represents a necessary lower network service for the wide diffusion of new multimedia network applications. Multicast transmission may use network resources more efficiently than multiple point-to-point messages; however, creating optimal multicast trees (Steiner Tree Problem in networks) is prohibitively expensive. This paper proposes a distributed algorithm for the heuristic solution of the Steiner Tree Problem, allowing the construction of effective distribution trees using a coordination protocol among the network nodes. Furthermore, we propose a novel distributed technique for dynamically updating the multicast tree. The approach proposed has been implemented and extensively tested both in simulation, and on experimental networks. Performance evaluation indicates that the distributed algorithm performs as well as the centralized version, providing good levels of convergence time and communication complexity.     

Solving a system of linear equations: From centralized to distributed algorithms

For a wide range of control engineering applications, the problem of solving a system of linear equations is often encountered and has been well studied. Traditionally, this problem has been mainly solved in a centralized manner. However, for applications related to large-scale complex networked systems, centralized algorithms are often subjected to some practical issues due to limited computational power and communication bandwidth. As a promising and viable alternative, distributed algorithms can effectively address the issues associated with centralized algorithms by solving the problem efficiently in a multi-agent setting that accords with the distributed nature of networked systems. Distributed algorithms decompose the entire problem into many sub-problems that are solved by individual agents in a cooperative manner. In this survey paper, we provide a detailed overview of the state of the art relevant to distributed algorithms for solving a system of linear equations. We will first review basic distributed algorithms including both discrete-time and continuous-time algorithms. Then we will discuss the extended algorithms to achieve communication efficiency. Furthermore, we will also introduce distributed algorithms to obtain the minimum-norm solution for a system of linear equations with multiple solutions, as well as the least-squares solution when there is no solution. Finally, the relationship of distributed algorithms for solving a system of linear equations to the existing distributed optimization algorithms is discussed.     

基于Web Service的分布式图书管理系统研究

随着高校的合并与扩招,许多的高校拥有多个校区和多个分布异地的图书馆,因此,如何实现各个分布式图书馆的集中管理,是一个典型的切实际的难题,本文从实际需求出发,分析了解决分布式问题的各种技术的优缺点,最后采用WebService技术解决高校多个分布式图书馆的集中管理问题。     

高阶Hopfield神经网求解合取范式可满足性问题

本文提出用高阶Ｈｏｐｆｉｅｌｄ神经网络求解ＳＡＴ问题，给出了连续及离散高阶神经网络模型与相应的离散快速求解算法，证明了网络的稳定性，并用实验证明了该方法的可行性，且将该算法与Ｌｏｃａｌ Ｓｅａｒｃｈ算法进行了比较。     

Performance Analyses of Nature-inspired Algorithms on the Traveling Salesman’s Problems for Strategic Management

This paper carries out a performance analysis of major Nature-inspired Algorithms in solving the benchmark symmetric and asymmetric Traveling Salesman’s Problems (TSP). Knowledge of the workings of the TSP is very useful in strategic management as it provides useful guidance to planners. After critical assessments of the performances of eleven algorithms consisting of two heuristics (Randomized Insertion Algorithm and the Honey Bee Mating Optimization for the Travelling Salesman’s Problem), two trajectory algorithms (Simulated Annealing and Evolutionary Simulated Annealing) and seven population-based optimization algorithms (Genetic Algorithm, Artificial Bee Colony, African Buffalo Optimization, Bat Algorithm, Particle Swarm Optimization, Ant Colony Optimization and Firefly Algorithm) in solving the 60 popular and complex benchmark symmetric Travelling Salesman’s optimization problems out of the total 118 as well as all the 18 asymmetric Travelling Salesman’s Problems test cases available in TSPLIB91. The study reveals that the African Buffalo Optimization and the Ant Colony Optimization are the best in solving the symmetric TSP, which is similar to intelligence gathering channel in the strategic management of big organizations, while the Randomized Insertion Algorithm holds the best promise in asymmetric TSP instances akin to strategic information exchange channels in strategic management.     

Solving weighted CSPs with meta-constraints by reformulation into satisfiability modulo theories

We introduce WSimply, a new framework for modelling and solving Weighted Constraint Satisfaction Problems (WCSP) using Satisfiability Modulo Theories (SMT) technology. In contrast to other well-known approaches designed for extensional representation of goods or no-goods, and with few declarative facilities, our approach aims to follow an intensional and declarative syntax style. In addition, our language has built-in support for some meta-constraints, such as priority and homogeneity, which allows the user to easily specify rich requirements on the desired solutions, such as preferences and fairness. We propose two alternative strategies for solving these WCSP instances using SMT. The first is the reformulation into Weighted SMT (WSMT) and the application of satisfiability test based algorithms from recent contributions in the Weighted Maximum Satisfiability field. The second one is the reformulation into an operation research-like style which involves an optimisation variable or objective function and the application of optimisation SMT solvers. We present experimental results of two well-known problems: the Nurse Rostering Problem (NRP) and a variant of the Balanced Academic Curriculum Problem (BACP), and provide some insights into the impact of the addition of meta-constraints on the quality of the solutions and the solving time.     

An ILP improvement procedure for the Open Vehicle Routing Problem

We address the Open Vehicle Routing Problem (OVRP), a variant of the “classical” (capacitated and distance constrained) Vehicle Routing Problem (VRP) in which the vehicles are not required to return to the depot after completing their service. We present a heuristic improvement procedure for OVRP based on Integer Linear Programming (ILP) techniques. Given an initial feasible solution to be possibly improved, the method follows a destruct-and-repair paradigm, where the given solution is randomly destroyed (i.e., customers are removed in a random way) and repaired by solving an ILP model, in the attempt of finding a new improved feasible solution. The overall procedure can be considered as a general framework which could be extended to cover other variants of Vehicle Routing Problems. We report computational results on benchmark instances from the literature. In several cases, the proposed algorithm is able to find the new best known solution for the considered instances.