João Pedro Pedroso

Hierarchical Clustering of Time-Series Data Streams

This paper presents and analyzes an incremental system for clustering streaming time series. The Online Divisive-Agglomerative Clustering (ODAC) system continuously maintains a tree-like hierarchy of clusters that evolves with data, using a top-down strategy. The splitting criterion is a correlation-based dissimilarity measure among time series, splitting each node by the farthest pair of streams. The system also uses a merge operator that reaggregates a previously split node in order to react to changes in the correlation structure between time series. The split and merge operators are triggered in response to changes in the diameters of existing clusters, assuming that in stationary environments, expanding the structure leads to a decrease in the diameters of the clusters. The system is designed to process thousands of data streams that flow at a high rate. The main features of the system include update time and memory consumption that do not depend on the number of examples in the stream. Moreover, the time and memory required to process an example decreases whenever the cluster structure expands. Experimental results on artificial and real data assess the processing qualities of the system, suggesting a competitive performance on clustering streaming time series, exploring also its ability to deal with concept drift.

Support vector machines with different norms: motivation, formulations and results

Abstract We introduce two formulations for training support vector machines, based on considering the L 1 and L ∞ norms instead of the currently used L 2 norm, and maximising the margin between the separating hyperplane and each data sets using L 1 and L ∞ distances. We exploit the geometrical properties of these different norms, and propose what kind of results should be expected for them. Formulations in mathematical programming for linear problems corresponding to L 1 and L ∞ norms are also provided, for both the separable and non-separable cases. We report results obtained for some standard benchmark problems, which confirmed that the performance of all the formulations is similar. As expected, the CPU time required for machines solvable with linear programming is much shorter.

Bin packing and related problems

We present an exact method, based on an arc-flow formulation with side constraints, for solving bin packing and cutting stock problems --- including multi-constraint variants --- by simply representing all the patterns in a very compact graph. Our method includes a graph compression algorithm that usually reduces the size of the underlying graph substantially without weakening the model. As opposed to our method, which provides strong models, conventional models are usually highly symmetric and provide very weak lower bounds. Our formulation is equivalent to Gilmore and Gomorys, thus providing a very strong linear relaxation. However, instead of using column-generation in an iterative process, the method constructs a graph, where paths from the source to the target node represent every valid packing pattern. The same method, without any problem-specific parameterization, was used to solve a large variety of instances from several different cutting and packing problems. In this paper, we deal with vector packing, graph coloring, bin packing, cutting stock, cardinality constrained bin packing, cutting stock with cutting knife limitation, cutting stock with binary patterns, bin packing with conflicts, and cutting stock with binary patterns and forbidden pairs. We report computational results obtained with many benchmark test data sets, all of them showing a large advantage of this formulation with respect to the traditional ones.

Tabu Search for Mixed Integer Programming

This paper introduces tabu search for the solution of general linear integer problems. Search is done on integer variables; if there are continuous variables, their corresponding value is determined through the solution of a linear program, which is also used to evaluate the integer solution. The complete tabu search procedure includes an intensification and diversification procedure, whose effects are analysed on a set of benchmark problems.

Heuristics and exact methods for number partitioning

Number partitioning is a classical NP-hard combinatorial optimization problem, whose solution is challenging for both exact and approximative methods. This work presents a new algorithm for number partitioning, based on ideas drawn from tree search, breadth first search, and beam search. A new set of benchmark instances for this problem is also proposed. The behavior of the new method on this and other testbeds is analyzed and compared to other well known heuristics and exact algorithms.

Maximizing Expectation on Vertex-Disjoint Cycle Packing

This paper proposes a method for computing the expectation for the length of a maximum set of vertex-disjoint cycles in a digraph where vertices and/or arcs are subject to failure with a known probability. This method has an immediate practical application: it can be used for the solution of a kidney exchange program in the common situation where the underlying graph is unreliable. Results for realistic benchmark instances are reported and analyzed.

Tree search for the stacking problem

The stacking problem is a hard combinatorial optimization problem with high practical interest in, for example, steel storage or container port operations. In this problem, a set of items is stored in a warehouse for a period of time, and a crane is used to place them in a limited number of stacks. Since the entrance and exit of items occurs in an arbitrary order, items may have to be relocated in order to reach and deliver other items below them. The objective of the problem is to find a feasible sequence of movements that delivers all items, while minimizing the total number of movements.We study the scalability of an exact approach to this problem, and propose two heuristic methods to solve it approximately. The two heuristic approaches are a multiple simulation algorithm using semi-greedy construction heuristics, and a stochastic best-first tree search algorithm. The two methods are compared in a set of challenging instances, revealing a superior performance of the tree search approach in most cases.

Heuristic search for the stacking problem

This paper presents the Stacking Problem, a hard combinatorial optimization problem concerning handling and storage of items in a warehouse, where they are handled by a crane and organized into stacks. We define the problem, study its complexity class, and present a mathematical programming model to solve it. In order to tackle medium- or large-scale instances, we propose a simulation-based algorithm using semi-greedy construction heuristics. This simple approach allows for multiple constructions, finding solutions within reasonable time even for large instances. Three semi-greedy heuristics are proposed and compared in an extensive computational experiment, where we study the relation between the number of constructions and the best solution obtained using each heuristic.

Solving the deterministic and stochastic uncapacitated facility location problem: from a heuristic to a simheuristic

Abstract The uncapacitated facility location problem (UFLP) is a popular combinatorial optimization problem with practical applications in different areas, from logistics to telecommunication networks. While most of the existing work in the literature focuses on minimizing total cost for the deterministic version of the problem, some degree of uncertainty (e.g., in the customers’ demands or in the service costs) should be expected in real-life applications. Accordingly, this paper proposes a simheuristic algorithm for solving the stochastic UFLP (SUFLP), where optimization goals other than the minimum expected cost can be considered. The development of this simheuristic is structured in three stages: (i) first, an extremely fast savings-based heuristic is introduced; (ii) next, the heuristic is integrated into a metaheuristic framework, and the resulting algorithm is tested against the optimal values for the UFLP; and (iii) finally, the algorithm is extended by integrating it with simulation techniques, and the resulting simheuristic is employed to solve the SUFLP. Some numerical experiments contribute to illustrate the potential uses of each of these solving methods, depending on the version of the problem (deterministic or stochastic) as well as on whether or not a real-time solution is required.

A branch-and-bound procedure for forest harvest scheduling problems addressing aspects of habitat availability

In the literature, the most widely referred approaches regarding forest harvesting scheduling problems involving environmental concerns have typically addressed constraints on the maximum clear-cut area. Nevertheless, the solutions arising from those approaches in general display a loss of habitat availability. Such loss endangers the survival of many wild species. This study presents a branch-and-bound procedure designed to find good feasible solutions, in a reasonable time, to forest harvest scheduling problems with constraints on the clear-cut area and habitat availability. Two measures are applied for the habitat availability constraints: the area of all habitats and the connectivity between them. In each branch of the branch-and-bound tree, a partial solution leads to two children nodes, corresponding to the cases of harvesting or not harvesting a given stand in a given period. Pruning is based on constraint violations or unreachable objective values. Computational results are reported.