A Gaussian Mixed Model for Learning Discrete Bayesian Networks

In this paper, we address the problem of learning discrete Bayesian networks from noisy data. A graphical model based on a mixture of Gaussian distributions with categorical mixing structure coming from a discrete Bayesian network is considered. The network learning is formulated as a maximum likelihood estimation problem and performed by employing an EM algorithm. The proposed approach is relevant to a variety of statistical problems for which Bayesian network models are suitable—from simple regression analysis to learning gene/protein regulatory networks from microarray data.     

Discrete Capacity Assignment in IP networks using Particle Swarm Optimization

This paper presents a design methodology for IP networks under end-to-end Quality-of-Service (QoS) constraints. Particularly, we consider a more realistic problem formulation in which the link capacities of a general-topology packet network are discrete variables. This Discrete Capacity Assignment (DCA) problem can be classified as a constrained combinatorial optimization problem. A refined TCP/IP traffic modeling technique is also considered in order to estimate performance metrics for networks loaded by realistic traffic patterns. We propose a discrete variable Particle Swarm Optimization (PSO) procedure to find solutions for the problem. A simple approach called Bottleneck Link Heuristic (BLH) is also proposed to obtain admissible solutions in a fast way. The PSO performance, compared to that one of an exhaustive search (ES) procedure, suggests that the PSO algorithm provides a quite efficient approach to obtain (near) optimal solutions with small computational effort.     

Designing WDM Optical Networks Using Branch-and-Price

In this paper, we present an exact solution procedure for the design of two-layer wavelength division multiplexing (WDM) optical networks with wavelength changers and bifurcated flows. This design problem closely resembles the traditional multicommodity flow problem, except that in the case of WDM optical networks, we are concerned with the routing of multiple commodities in two network layers. Consequently, the corresponding optimization models have to deal with two types of multicommodity variables defined for each of the network layers. The proposed procedure represents one of the first branch-and-price algorithms for a general WDM optical network setting with no assumptions on the number of logical links that can be established between nodes in the network. We apply our procedure in a computational study with four different network configurations. Our results show that for the three tested network configurations our branch-and-price algorithm provides solutions that are on average less than 5 % from optimality. We also provide a comparison of our branch-and-price algorithm with two simple variants of the upper bounding heuristic procedure HLDA that is commonly used for WDM optical network design.     

Locating the two-median of a tree network with continuous link demands

Typical formulations of thep-median problem on a network assume discrete nodal demands. However, for many problems, demands are better represented by continuous functions along the links, in addition to nodal demands. For such problems, optimal server locations need not occur at nodes, so that algorithms of the kind developed for the discrete demand case can not be used. In this paper we show how the 2-median of a tree network with continuous link demands can be found using an algorithm based on sequential location and allocation. We show that the algorithm will converge to a local minimum and then present a procedure for finding the global minimum solution.     

Development of the tree-based link labeling algorithm for optimal path-finding in urban transportation networks

The optimal path-finding algorithm which is an important module in developing route guidance systems and traffic control systems has to provide correct paths to consider U-turns, P-turns, and no-left-turns in urban transportation networks.Traditional methods which have been used to consider those regulations on urban transportation networks can be categorized into network representation and algorithmic methods like the vine-building algorithm. First, network representation methods use traditional optimal path-finding algorithms with modifications to the network structure: for example, just adding dummy nodes and links to the existing network allows constraint-search in the network. This method which creates large networks is hard to implement and introduces considerable difficulties in network coding. With the increased number of nodes and links, the memory requirement tremendously increases, which causes the processing speed to slow down. For these reasons, the method has not been widely accepted for incorporating turning regulations in optimal path-finding problems in transportation networks. Second, algorithmic methods, as they are mainly based on the vine-building algorithm, have been suggested for determining optimal path for networks with turn penalties and prohibitions. However, the algorithms, although they nicely reflect the characteristics of urban transportation networks, frequently provide infeasible or suboptimal solutions.The algorithm to be suggested in this research is a method which is basically based on Dijkstra's algorithm [1] and the tree-building algorithm used to construct optimal paths. Unlike the traditional node labeling algorithms which label each node with minimum estimated cost, this algorithm labels each link with minimum estimated cost.Comparison with the vine-building algorithm shows that the solution of the link-labeling algorithm is better than that of the vine-building algorithm which very frequently provides suboptimal solutions. As a result, the algorithm allows turning regulations, while providing an optimal solution within a reasonable time limit.     

Experimental evaluation of simulated annealing algorithms for the time-cost trade-off problem

This paper evaluates variants of a simulated annealing algorithm which solve the total cost minimization problem in activity networks in the case that discrete time-cost execution modes are allowed on the project activities. This problem is a special case of the well known discrete time-cost trade-off problem (DTCTP). Based on a sample of randomly generated activity networks, formal tests of statistical significance are utilized to test both the quality of solutions and the time efficiency of algorithms versus problem factors. A procedure issued from the extreme values statistics is also applied on problem instances in order to determine, on the one hand, the confidence interval estimate of the optimum solution for each algorithm and, on the other hand, when to stop the running of an algorithm.     

New computational results on the discrete time/cost trade-off problem in project networks

We describe a new exact procedure for the discrete time/cost trade-off problem in deterministic activity-on-the-arc networks of the CPM type, where the duration of each activity is a discrete, nonincreasing function of the amount of a single resource (money) committed to it. The objective is to construct the complete and efficient time/cost profile over the set of feasible project durations. The procedure uses a horizon-varying approach based on the iterative optimal solution of the problem of minimising the sum of the resource use over all activities subject to the activity precedence constraints and a project deadline. This optimal solution is derived using a branch-and-bound procedure which computes lower bounds by making convex piecewise linear underestimations of the discrete time/cost trade-off curves of the activities to be used as input for an adapted version of the Fulkerson labelling algorithm for the linear time/cost trade-off problem. Branching involves the selection of an activity in order to partition its set of execution modes into two subsets which are used to derive improved convex piecewise linear underestimations. The procedure has been programmed in Visual C ++ under Windows NT and has been validated using a factorial experiment on a large set of randomly generated problem instances.     

Optimum synthesis of discrete capacitated networks with multi-terminal commodity flow requirements

Network design problems arise in a wide range of applied areas including telecommunications, computer networks, and transportation. In this paper, we address the following discrete capacitated multi-terminal network design problem. Given a connected digraph G = (V,A), a set of L potential facilities to be installed on each arc, and a set of K multi-terminal (non-simultaneous) commodity flow requirements, the problem is to find a set of facilities to install in order to route the K nonsimultaneous flows while minimizing the total fixed plus variable costs. We describe an exact procedure for solving this problem based on Benders decomposition. Our algorithm includes several features that significantly improve the efficiency of the basic approach. Computational results attest to the efficacy of the proposed algorithm, which can solve medium- to large-scale problems to optimality.     

Improving solution of discrete competitive facility location problems

We consider discrete competitive facility location problems in this paper. Such problems could be viewed as a search of nodes in a network, composed of candidate and customer demand nodes, which connections correspond to attractiveness between customers and facilities located at the candidate nodes. The number of customers is usually very large. For some models of customer behavior exact solution approaches could be used. However, for other models and/or when the size of problem is too high to solve exactly, heuristic algorithms may be used. The solution of discrete competitive facility location problems using genetic algorithms is considered in this paper. The new strategies for dynamic adjustment of some parameters of genetic algorithm, such as probabilities for the crossover and mutation operations are proposed and applied to improve the canonical genetic algorithm. The algorithm is also specially adopted to solve discrete competitive facility location problems by proposing a strategy for selection of the most promising values of the variables in the mutation procedure. The developed genetic algorithm is demonstrated by solving instances of competitive facility location problems for an entering firm.     

经济时间序列的连续参数小波网络预测模型

本文利用连续小波变换方法,给出一种连续参数小波网络。网络参数的学习采用一种类似神经网络的后向传播学习算法的随机梯度算法。另外,提出了一种借助小波分析理论指导网络参数赋初值的方法。进一步,通过对中国进出口贸易额时间序列预测建模的研究和仿真预测,提出了用连续参数小波网络建立经济时间序列预测模型的一般步骤和方法。预测结果表明,此模型具有较好的泛化、学习能力,可以有效地在数值上逼近时间序列难以定量描述的相互关系,所以利用连续参数小波网络建立的时间序列预测模型有较高的预测精度。     

Combination of fuzzy ranking and simulated annealing to improve discrete fracture inversion

Mathematical and computational modelling of discrete fracture networks is critical for the exploration and development of natural resource reservoirs. Utilizing the concept of fuzzy memberships, this paper advances the fundamental understanding in fracture network inversion and presents a systematic procedure to solve the most important problem in global optimization (simulated annealing): objective function formulation. First, a comprehensive field study identifies all potential components of an objective function. The components are statistical, geostatistical, mathematical and spatial measurements of fracture properties (location, orientation and size). The characteristic measurements can be input in parametric or non-parametric, discrete or continuum forms. Next, sensitivity analysis and fuzzy logic are combined to rank the candidate components based on their effects on the final objective function value and optimization convergence. The process negates guess works in objective function formulation by automatic selection of highly ranked components and their corresponding weighting factors. A case study is applied to a surface DFN in New York. The derived discrete fracture network is representative of the field data.     

Tabu Search for Frequency Assignment in Mobile Radio Networks

The main goal of the Frequency Assignment Problem in mobile radio networks consists of assigning a limited number of frequencies to each radio cell in a cellular network while minimizing electromagnetic interference due to the reuse of frequencies. This problem, known to be NP-hard, is of great importance in practice since better solutions will allow a telecommunications operator to manage larger cellular networks. This paper presents a new Tabu Search algorithm for this application. The algorithm is tested on realistic and large problem instances and compared with other methods based on simulated annealing, constraint programming and graph coloring algorithms. Empirical evidence shows that the Tabu algorithm is very competitive by giving the best solutions to the tested instances.     

Generalized loopy 2U: A new algorithm for approximate inference in credal networks

Credal networks generalize Bayesian networks by relaxing the requirement of precision of probabilities. Credal networks are considerably more expressive than Bayesian networks, but this makes belief updating NP-hard even on polytrees. We develop a new efficient algorithm for approximate belief updating in credal networks. The algorithm is based on an important representation result we prove for general credal networks: that any credal network can be equivalently reformulated as a credal network with binary variables; moreover, the transformation, which is considerably more complex than in the Bayesian case, can be implemented in polynomial time. The equivalent binary credal network is then updated by L2U, a loopy approximate algorithm for binary credal networks. Overall, we generalize L2U to non-binary credal networks, obtaining a scalable algorithm for the general case, which is approximate only because of its loopy nature. The accuracy of the inferences with respect to other state-of-the-art algorithms is evaluated by extensive numerical tests.     

A modified active set algorithm for transportation discrete network design bi-level problem

Transportation discrete network design problem (DNDP) is about how to modify an existing network of roads and highways in order to improve its total system travel time, and the candidate road building or expansion plan can only be added as a whole. DNDP can be formulated into a bi-level problem with binary variables. An active set algorithm has been proposed to solve the bi-level discrete network design problem, while it made an assumption that the capacity increase and construction cost of each road are based on the number of lanes. This paper considers a more general case when the capacity increase and construction cost are specified for each candidate plan. This paper also uses numerical methods instead of solvers to solve each step, so it provides a more direct understanding and control of the algorithm and running procedure. By analyzing the differences and getting corresponding solving methods, a modified active set algorithm is proposed in the paper. In the implementation of the algorithm and the validation, we use binary numeral system and ternary numeral system to avoid too many layers of loop and save storage space. Numerical experiments show the correctness and efficiency of the proposed modified active set algorithm.     

Virtual path layouts optimizing total hop count on ATM tree networks

This paper presents some algorithmic results concerning virtual path layouts for the one-to-many communication problem in ATM tree networks. The ATM network model is based on covering the network with a layout of virtual paths, under some constraints on the allowed load, namely, the number of paths that can share an edge. The quality measure used is the hop count, namely, the number of edges traversed between two vertices that need to communicate. Whereas most former results concerned the maximum hop count of the virtual path layout, our interest here is in measuring its total hop count, or alternatively its average hop count. The paper presents a dynamic programming algorithm for planning ATM network layouts with minimal total hop count for one-to-many requirements under load constraints over the class of tree networks.     

A heuristic method for the minimum toll booth problem

This paper addresses the toll pricing problem in which the objective is to minimize the number of required toll facilities in a traffic network. The problem is shown to be NP-hard. To obtain a solution in a reasonable time, an effective metaheuristic algorithm is developed. The algorithm uses a local search technique in which the neighborhood function employs the dynamic slope scaling procedure to deal with the fixed charge nature of the objective function. Numerical results from 50 randomly generated and three real networks are reported.     

Isodistant points in competitive network facility location

An isodistant point is any point on a network which is located at a predetermined distance from some node. For some competitive facility location problems on a network, it is verified that optimal (or near-optimal) locations are found in the set of nodes and isodistant points (or points in the vicinity of isodistant points). While the nodes are known, the isodistant points have to be determined for each problem. Surprisingly, no algorithm has been proposed to generate the isodistant points on a network. In this paper, we present a variety of such problems and propose an algorithm to find all isodistant points for given threshold distances associated with the nodes. The number of isodistant points is upper bounded by nm, where n and m are the number of nodes and the number of edges, respectively. Computational experiments are presented which show that isodistant points can be generated in short run time and the number of such points is much smaller than nm. Thus, for networks of moderate size, it is possible to find optimal (or near-optimal) solutions through the Integer Linear Programming formulations corresponding to the discrete version of such problems, in which a finite set of points are taken as location candidates.     

New formulations and solution procedures for the hop constrained network design problem

Optical fiber provides tremendous advantages in being able to carry a wide range of services including video on demand, video conferencing, distance learning, remote medical imaging, and telecommuting. The high capacities encourage carriers to create networks that are substantially sparser than previous copper based networks. A recent publication by the Telecommunications Industry Association indicated that investment in fiber optics is projected to reach $35 billion in the year 2003. Given the magnitude of investments, the design of networks becomes a very important issue. Most telecommunication companies (telcos), IT consulting companies, network equipment manufacturers and network service providers have extensive network design groups. The primary function of these groups is to design the most efficient networks both in terms of costs and performance and maintain them. These designers need flexible tools to support topological network design decisions. These decisions involve significant levels of investments in transmissions and switching facilities, and impact the resulting networks’ performance fundamentally.In this paper we study a special type of a network design problem called the hop constrained backbone network design problem. We present new mathematical programming formulations of the problem and develop an efficient solution procedure based on the linear programming relaxation. Extensive computational results across a number of networks are reported.     

Metaheuristic hybridizations for the regenerator placement and dimensioning problem in sub-wavelength switching optical networks

Physical layer impairments severely limit the reach and capacity of optical systems, thereby hampering the deployment of transparent optical networks (i.e., no electrical signal regenerators are required). Besides, the high cost and power-consumption of regeneration devices makes it unaffordable for network operators to consider the opaque architecture (i.e., regeneration is available at every network node). In this context, translucent architectures (i.e., regeneration is only available at selected nodes) have emerged as the most promising short term solution to decrease costs and energy consumption in optical backbone networks. Concurrently, the coarse granularity and inflexibility of legacy optical technologies have re-fostered great interest in sub-wavelength switching optical networks, which introduce optical switching in the time domain so as to further improve resources utilization. In these networks, the complex regenerator placement and dimensioning problem emerges. In short, this problem aims at minimizing the number of electrical regenerators deployed in the network. To tackle it, in this paper both a greedy randomized adaptive search procedure and a biased random-key genetic algorithm are developed. Further, we enhance their performance by introducing both path-relinking and variable neighborhood descent as effective intensification procedures. The resulting hybridizations are compared among each other as well as against results from optimal and heuristic mixed integer linear programming formulations. Illustrative results over a broad range of network scenarios show that the biased random-key genetic algorithm working in conjunction with these two intensification mechanisms represents a compelling network planning algorithm for the design of future sub-wavelength optical networks.     

General Variable Neighborhood Search for computing graph separators

Computing graph separators in networks has a wide range of real-world applications. For instance, in telecommunication networks, a separator determines the capacity and brittleness of the network. In the field of graph algorithms, the computation of balanced small-sized separators is very useful, especially for divide-and-conquer algorithms. In bioinformatics and computational biology, separators are required in grid graphs providing a simplified representation of proteins. This papers presents a new heuristic algorithm based on the Variable Neighborhood Search methodology for computing vertex separators. We compare our procedure with the state-of-the-art methods. Computational results show that our procedure obtains the optimum solution in all of the small and medium instances, and high-quality results in large instances.