并行机间歇过程生产调度的遗传局部搜索算法

研究了一类集成分批的并行机间歇过程调度问题(parallel machine batch process scheduling problem,简称PBPSP),将此问题转化为固定费用运输问题(6xed charge transportation problem,简称FCTP)后,提出了具有集中邻域搜索机制和局部最优逃逸机制的遗传局部搜索算法(genetic local search algorithm,简称GLSA).GLSA算法用先根遍历边排列模式编码生成树解,具有高效的子树补充式单点交叉操作.将基于网络单纯型方法的邻域搜索作为变异算子,并提出了连续随机节点邻域搜索的集中邻域搜索策略以及随机旋转变异与全局邻域搜索相结合的局部最优逃逸策略,极大地强化了遗传局部搜索算法的全局寻优能力.实验表明:GLSA算法获得的解质量优于基于排列编码的遗传算法和基于矩阵编码的遗传算法,得到了所有Benchmark问题的最优解,且具有高鲁棒性.针对一定规模的FCTP问题,GLSA算法比Tabu启发式搜索算法具有更高的获得最优解几率.     

A refined search tree technique for Dominating Set on planar graphs

We establish a refined search tree technique for the parameterized DOMINATING SET problem on planar graphs. Here, we are given an undirected graph and we ask for a set of at most k vertices such that every other vertex has at least one neighbor in this set. We describe algorithms with running times O(8^kn) and O(8^kk+n³), where n is the number of vertices in the graph, based on bounded search trees. We describe a set of polynomial time data-reduction rules for a more general “annotated” problem on black/white graphs that asks for a set of k vertices (black or white) that dominate all the black vertices. An intricate argument based on the Euler formula then establishes an efficient branching strategy for reduced inputs to this problem. In addition, we give a family examples showing that the bound of the branching theorem is optimal with respect to our reduction rules. Our final search tree algorithm is easy to implement; its analysis, however, is involved.     

Pseudo-Kernelization: A Branch-then-Reduce Approach for FPT Problems

Pseudo-kernelization is introduced in this paper as a new strategy for improving fixed-parameter algorithms. This new technique works for bounded search tree algorithms by identifying favorable branching conditions whose absence could be used to reduce the size of corresponding problem instances. Pseudo-kernelization applies well to hitting set problems. It can be used either to improve the search tree size of a 3-Hitting-Set algorithm from O^*(2.179^k) to O^*(2.05^k), or to improve the kernel size from k³ to 27k. In this paper the parameterized 3-Hitting-Set and Face Cover problems are used as typical examples.     

A CSP Search Algorithm with Responsibility Sets and Kernels

A CSP search algorithm, like FC or MAC, explores a search tree during its run. Every node of the search tree can be associated with a CSP created by the refined domains of unassigned variables. If the algorithm detects that the CSP associated with a node is insoluble, the node becomes a dead-end. A strategy of pruning “by analogy” states that the current node of the search tree can be discarded if the CSP associated with it is “more constrained” than a CSP associated with some dead-end node. In this paper we present a method of pruning based on the above strategy. The information about the CSPs associated with dead-end nodes is kept in the structures called responsibility sets and kernels. We term the method that uses these structures for pruning RKP, which is abbreviation of Responsibility set, Kernel, Propagation. We combine the pruning method with algorithms FC and MAC. We call the resulting solvers FC-RKP and MAC-RKP, respectively. Experimental evaluation shows that MAC-RKP outperforms MAC-CBJ on random CSPs and on random graph coloring problems. The RKP-method also has theoretical interest. We show that under certain restrictions FC-RKP simulates FC-CBJ. It follows from the fact that intelligent backtracking implicitly uses the strategy of pruning “by analogy.”     

Benders decomposition with integer subproblem

The application of Benders decomposition method to a problem might result in a subproblem including integer variables. In this case, it is not able to apply the classical Benders algorithm. In this study we present a Branch-and-Cut algorithm, which introduces the notion of “Local Cuts” as well as “Global Cuts”. The integrality constraints of the subproblem are relaxed and the relaxed problem is solved in a branch-and-bound framework, where in each node, the Benders algorithm is applied between the master problem and the relaxed subproblem. Benders cuts generated in a node of the branch-and bound tree are proved to be valid for all its descendants, but they are not necessarily valid for the non-descendant nodes. These cuts, referred to as local cuts, can be used to warm start the master problem of each descendant node, thus leading to better initial bounds. Furthermore, a novel way is presented for defining the local cuts in a general form. This general form is in fact a function of the subproblems’ variables and enables us to reuse the generated (local) cuts in the whole tree by updating some values of the function. The performance of the proposed algorithm is tested on the classical Capacitated Fixed Charge Multiple Knapsack Problem (CFCMKP).     

Distributed coordinate descent for generalized linear models with regularization

Generalized linear model with L ₁ and L ₂ regularization is a widely used technique for solving classification, class probability estimation and regression problems. With the numbers of both features and examples growing rapidly in the fields like text mining and clickstream data analysis parallelization and the use of cluster architectures becomes important. We present a novel algorithm for fitting regularized generalized linear models in the distributed environment. The algorithm splits data between nodes by features, uses coordinate descent on each node and line search to merge results globally. Convergence proof is provided. A modifications of the algorithm addresses slow node problem. For an important particular case of logistic regression we empirically compare our program with several state-of-the art approaches that rely on different algorithmic and data spitting methods. Experiments demonstrate that our approach is scalable and superior when training on large and sparse datasets.     

The air-cargo consolidation problem with pivot weight: Models and solution methods

In the international air cargo business, freight is usually consolidated into containers, called Unit Load Devices (ULDs). The transportation charge of a ULD depends on whether the total weight exceeds a certain threshold, called the pivot weight. If the total load tendered by a freight forwarder is less than the pivot weight, it gets charged at the under-pivot rate. Any portion of the load that exceeds the pivot weight is charged at the over-pivot rate. This scheme is adopted for safety reasons to avoid the overloading of ULDs. We formulate a mixed integer program, and propose four solution methodologies for the air-cargo consolidation problem under the pivot-weight scheme (ACPW). These are exact solution approaches based on branch-and-price, a best-fit decreasing loading heuristic, and two extended local branching heuristics (a multi-local tree search and a relaxation-induced neighborhood search). The local branching heuristic with relaxation-induced neighborhood search is found to outperform other approaches in terms of solution quality and computational time. Problems with up to 400 shipments and 80 containers are solved to within 3.4% of optimality in less than 20 min.     

Relaxed Balance Using Standard Rotations

Abstract. In search trees with relaxed balance, rebalancing transformations need not be connected with updates, but may be delayed. For standard AVL tree rebalancing, we prove that even though the rebalancing operations are uncoupled from updates, their total number is bounded by O(M log (M+N)) , where M is the number of updates to an AVL tree of initial size N . Hence, relaxed balancing of AVL trees comes at no extra cost asymptotically. Furthermore, our scheme differs from most other relaxed balancing schemes in an important aspect: No rebalancing transformation can be done in the wrong direction, i.e., no performed rotation can make the tree less balanced. Moreover, each performed rotation indeed corresponds to a real imbalance situation in the tree. Finally, and perhaps most importantly, our structure is capable of forgetting registered imbalance if later updates happen to improve the situation. Our results are of theoretical interest and have possible sequential and parallel applications.     

Approximating Optimal Binary Decision Trees

We give a (ln?n+1)-approximation for the decision tree (DT) problem. An instance of DT is a set of m binary tests T=(T ₁,…,T _m ) and a set of n items X=(X ₁,…,X _n ). The goal is to output a binary tree where each internal node is a test, each leaf is an item and the total external path length of the tree is minimized. Total external path length is the sum of the depths of all the leaves in the tree. DT has a long history in computer science with applications ranging from medical diagnosis to experiment design. It also generalizes the problem of finding optimal average-case search strategies in partially ordered sets which includes several alphabetic tree problems. Our work decreases the previous best upper bound on the approximation ratio by a constant factor. We provide a new analysis of the greedy algorithm that uses a simple accounting scheme to spread the cost of a tree among pairs of items split at a particular node. We conclude by showing that our upper bound also holds for the DT problem with weighted tests.     

Independence of Containing Patterns Property and Its Application in Tree Pattern Query Rewriting Using Views

We show that several classes of tree patterns observe the independence of containing patterns property, that is, if a pattern is contained in the union of several patterns, then it is contained in one of them. We apply this property to two related problems on tree pattern rewriting using views. First, given view V and query Q, is it possible for Q to have an equivalent rewriting using V which is the union of two or more tree patterns, but not an equivalent rewriting which is a single pattern? This problem is of both theoretical and practical importance because, if the answer is no, then, to find an equivalent rewriting of a tree pattern using a view, we should use more efficient methods, such as the polynomial time algorithm of Xu and Özsoyoglu (2005), rather than try to find the union of all contained rewritings (which takes exponential time in the worst case) and test its equivalence to Q. Second, given a set S of views, we want to know under what conditions a subset S′ of S is redundant in the sense that for any query Q, the contained rewritings of Q using the views in S′ are contained in those using the views in S???S′. Solving this problem can help us to, for example, choose the minimum number of views to be cached, or better design the virtual schema in a mediated data integration system, or avoid repeated calculation in query optimization. For the first problem, we identify several classes of tree patterns for which the equivalent rewriting can be expressed as a single tree pattern. For the second problem, we present necessary and sufficient conditions for S′ to be redundant with respect to some classes of tree patterns. For both problems we consider extension to cases where there are rewritings using the intersection of multiple views and/or where a schema graph is present.     

Fast k-nearest neighbors search using modified principal axis search tree

The problem of k-nearest neighbors (kNN) is to find the nearest k neighbors for a query point from a given data set. Among available methods, the principal axis search tree (PAT) algorithm always has good performance on finding nearest k neighbors using the PAT structure and a node elimination criterion. In this paper, a novel kNN search algorithm is proposed. The proposed algorithm stores projection values for all data points in leaf nodes. If a leaf node in the PAT cannot be rejected by the node elimination criterion, data points in the leaf node are further checked using their pre-stored projection values to reject more impossible data points. Experimental results show that the proposed method can effectively reduce the number of distance calculations and computation time for the PAT algorithm, especially for the data set with a large dimension or for a search tree with large number of data points in a leaf node.     

An introduction to nonlinear programming—III: Search procedures for unconstrained minimization

Attention is directed to the problem of numerically determining the minimum of a function l when there are no constraints on the minimizing variables y. Assuming that l is differentiable, the minimization problem can be regarded as being essentially equivalent to the problem of solving a system of n equations involving n unknowns (i.e. ?l/?m = 0). Additionally, each solution of the equations must be tested to establish the nature of the resulting stationary point (e.g. whether it provides a minimum). There has been considerable work done on these types of problems and many algorithms have been proposed for their solution. This paper presents a discussion and statement of some of these algorithms. Both direct search (i.e. procedures that do not use derivative information) and gradient search methods are discussed and compared. While unconstrained minimization problems are themselves important, the algorithms used for their solution often provide the basis for the numerical solution of constrained optimization problems. Thus, this discussion, as does the discussion of Parts I and II, is germane to the consideration in Part IV of solution algorithms for constrained optimization problems.     

用回溯法求哈密顿通路

回溯法是一种按照深度优先的策略从根结点开始搜索解空间树的算法，该算法可以用来求出问题的全部解，也可以在求出问题的一个解之后停止对问题的求解，即只求该问题是否有解。哈密顿通路就是判断图中是否存在一条通过所有顶点一次且仅一次的路径。本文主要讲的就是用回溯法来求解一个任意的图中是否存在一条哈密顿通路的问题，并用具体的算法来实现它。     

Computationally efficient solution of the multi-item, capacitated lot-sizing problem

The problem of multi-item, single-level, dynamic lotsizing in the presence of a single capacitated resource is addressed. A model based on variable redefinition is developed leading to a solution strategy based on a branch-and-bound search with sharp low bounds. The multi-item low bound problems are solved by column generation with the capacity constraints as the linking constraints. The resulting subproblems separate into as many single-item, uncapacitated lotsizing problems as there are items. These subproblems are solved as shortest-path problems. Good upper bounds are also generated by solving an appropriate minimum-cost network flow problem at each node of the branch-and-bound tree. The resulting solution scheme is very efficient in terms of computation time. Its efficiency is demonstrated by computational testing on a ste of benchmark problem instances and is attributable to the sharpness of the lower bounds, the efficiency with which the low bound problems are solved and the frequent generation of good upper bounds; all of which leading to a high exclusion rate.     

非平衡二叉树多类支持向量机分类方法

提出一种新的基于非平衡二叉树的支持向量机多类别分类方法。该方法通过分析已知类别样本的先验分布知识,构造一个二叉决策树,使容易区分的类别从根节点开始逐层分割出来,以获得较高的推广能力。该方法解决了传统分类算法中所存在的不可分区域问题,在训练时只需构造N-1个SVM分类器,而测试时的判决次数小于N。将该方法应用于人脸识别实验。测试结果表明,与传统分类算法相比,该方法的平均分类时间是最少的。     

Lower tolerance-based Branch and Bound algorithms for the ATSP

In this paper, we develop a new tolerance-based Branch and Bound algorithm for solving NP-hard problems. In particular, we consider the asymmetric traveling salesman problem (ATSP), an NP-hard problem with large practical relevance. The main algorithmic contribution is our lower bounding strategy that uses the expected costs of including arcs in the solution to the assignment problem relaxation of the ATSP, the so-called lower tolerance values. The computation of the lower bound requires the calculation of a large set of lower tolerances. We apply and adapt a finding from [23] that makes it possible to compute all lower tolerance values efficiently. Computational results show that our Branch and Bound algorithm exhibits very good performance in comparison with state-of-the-art algorithms, in particular for difficult clustered ATSP instances.     

On a stochastic sequencing and scheduling problem

We present a framework for solving multistage pure 0–1 programs for a widely used sequencing and scheduling problem with uncertainty in the objective function coefficients, the constraint matrix and the right-hand side. The problem has the following form: given a set of operations to be executed along a time horizon, find a schedule to minimize a function included by the expected operations cost over the scenarios under consideration, subject to a set of constraints. Typical elements are: limited availability of the resources, multiperiod operations, subsets of operations with exclusivity and implicative constraints, precedence relationships in the execution of the operations, etc. The stochasticity is in the resources’ consumption by the operations, their availability and the operations cost along the time horizon. A multistage scenario analysis with complete recourse is used. Given the high dimensions of the problem and its combinatorial nature, it is not realistic to obtain the optimal solution for the problem. Instead, we present the so-called Fix-and-Relax Coordination algorithmic framework to exploit the characteristics of the non-anticipativity constraints for each scenario group in the stochastic model. This exploitation basically consists of selectively exploring the nodes of the branching trees in which the branch-and-bound tree is decomposed while the non-anticipativity constraints are relaxed. The algorithm is specifically designed for coordinating and reinforcing the node pruning, and the branching node and variable selection at each branching tree, such that the non-anticipativity constraints are satisfied. Some computational experience is reported.     

Variable neighborhood search for the cost constrained minimum label spanning tree and label constrained minimum spanning tree problems

Given an undirected graph whose edges are labeled or colored, edge weights indicating the cost of an edge, and a positive budget B, the goal of the cost constrained minimum label spanning tree (CCMLST) problem is to find a spanning tree that uses the minimum number of labels while ensuring its cost does not exceed B. The label constrained minimum spanning tree (LCMST) problem is closely related to the CCMLST problem. Here, we are given a threshold K on the number of labels. The goal is to find a minimum weight spanning tree that uses at most K distinct labels. Both of these problems are motivated from the design of telecommunication networks and are known to be NP-complete [15].In this paper, we present a variable neighborhood search (VNS) algorithm for the CCMLST problem. The VNS algorithm uses neighborhoods defined on the labels. We also adapt the VNS algorithm to the LCMST problem. We then test the VNS algorithm on existing data sets as well as a large-scale dataset based on TSPLIB [12] instances ranging in size from 500 to 1000 nodes. For the LCMST problem, we compare the VNS procedure to a genetic algorithm (GA) and two local search procedures suggested in [15]. For the CCMLST problem, the procedures suggested in [15] can be applied by means of a binary search procedure. Consequently, we compared our VNS algorithm to the GA and two local search procedures suggested in [15]. The overall results demonstrate that the proposed VNS algorithm is of high quality and computes solutions rapidly. On our test datasets, it obtains the optimal solution in all instances for which the optimal solution is known. Further, it significantly outperforms the GA and two local search procedures described in [15].     

On the solution of mixed-integer nonlinear programming models for computer aided molecular design

This paper addresses the efficient solution of computer aided molecular design (CAMD) problems, which have been posed as mixed-integer nonlinear programming models. The models of interest are those in which the number of linear constraints far exceeds the number of nonlinear constraints, and with most variables participating in the nonconvex terms. As a result global optimization methods are needed. A branch-and-bound algorithm (BB) is proposed that is specifically tailored to solving such problems. In a conventional BB algorithm, branching is performed on all the search variables that appear in the nonlinear terms. This translates to a large number of node traversals. To overcome this problem, we have proposed a new strategy for branching on a set of linear branchingfunctions, which depend linearly on the search variables. This leads to a significant reduction in the dimensionality of the search space. The construction of linear underestimators for a class of functions is also presented. The CAMD problem that is considered is the design of optimal solvents to be used as cleaning agents in lithographic printing.