Near-optimal feature selection for large databases

We analyse a new optimization-based approach for feature selection that uses the nested partitions method for combinatorial optimization as a heuristic search procedure to identify good feature subsets. In particular, we show how to improve the performance of the nested partitions method using random sampling of instances. The new approach uses a two-stage sampling scheme that determines the required sample size to guarantee convergence to a near-optimal solution. This approach therefore also has attractive theoretical characteristics. In particular, when the algorithm terminates in finite time, rigorous statements can be made concerning the quality of the final feature subset. Numerical results are reported to illustrate the key results, and show that the new approach is considerably faster than the original nested partitions method and other feature selection methods.     

Structural Optimization with FEM-based Shakedown Analyses

In this paper, a mathematical programming formulation is presented for the structural optimization with respect to the shakedown analysis of 3-D perfectly plastic structures on basis of a finite element discretization. A new direct algorithm using plastic sensitivities is employed in solving this optimization formulation. The numerical procedure has been applied to carry out the shakedown analysis of pipe junctions under multi-loading systems. The new approach is compared to so-called derivative-free direct search methods. The computational effort of the proposed method is much lower compared to this methods.     

Approximation of the joint spectral radius using sum of squares

We provide an asymptotically tight, computationally efficient approximation of the joint spectral radius of a set of matrices using sum of squares (SOS) programming. The approach is based on a search for an SOS polynomial that proves simultaneous contractibility of a finite set of matrices. We provide a bound on the quality of the approximation that unifies several earlier results and is independent of the number of matrices. Additionally, we present a comparison between our approximation scheme and earlier techniques, including the use of common quadratic Lyapunov functions and a method based on matrix liftings. Theoretical results and numerical investigations show that our approach yields tighter approximations.     

Weighted quadrature rules for finite element methods

We discuss the numerical integration of polynomials times non-polynomial weighting functions in two dimensions arising from multiscale finite element computations. The proposed quadrature rules are significantly more accurate than standard quadratures and are better suited to existing finite element codes than formulas computed by symbolic integration. We validate this approach by introducing the new quadrature formulas into a multiscale finite element method for the two-dimensional reaction–diffusion equation.     

A nonmonotone trust-region line search method for large-scale unconstrained optimization

We consider an efficient trust-region framework which employs a new nonmonotone line search technique for unconstrained optimization problems. Unlike the traditional nonmonotone trust-region method, our proposed algorithm avoids resolving the subproblem whenever a trial step is rejected. Instead, it performs a nonmonotone Armijo-type line search in direction of the rejected trial step to construct a new point. Theoretical analysis indicates that the new approach preserves the global convergence to the first-order critical points under classical assumptions. Moreover, superlinear and quadratic convergence are established under suitable conditions. Numerical experiments show the efficiency and effectiveness of the proposed approach for solving unconstrained optimization problems.     

The convergence of conjugate gradient method with nonmonotone line search

The conjugate gradient method is a useful and powerful approach for solving large-scale minimization problems. Liu and Storey developed a conjugate gradient method, which has good numerical performance but no global convergence under traditional line searches such as Armijo line search, Wolfe line search, and Goldstein line search. In this paper we propose a new nonmonotone line search for Liu-Storey conjugate gradient method (LS in short). The new nonmonotone line search can guarantee the global convergence of LS method and has a good numerical performance. By estimating the Lipschitz constant of the derivative of objective functions in the new nonmonotone line search, we can find an adequate step size and substantially decrease the number of functional evaluations at each iteration. Numerical results show that the new approach is effective in practical computation.     

Multiple-retrieval case-based reasoning for course timetabling problems

The structured representation of cases by attribute graphs in a case-based reasoning (CBR) system for course timetabling has been the subject of previous research by the authors. In that system, the case base is organized as a decision tree and the retrieval process chooses those cases that are sub-attribute graph isomorphic to the new case. The drawback of that approach is that it is not suitable for solving large problems. This paper presents a multiple-retrieval approach that partitions a large problem into small solvable sub-problems by recursively inputting the unsolved part of the graph into the decision tree for retrieval. The adaptation combines the retrieved partial solutions of all the partitioned sub-problems and employs a graph heuristic method to construct the whole solution for the new case. We present a methodology which is not dependent upon problem-specific information and which, as such, represents an approach which underpins the goal of building more general timetabling systems. We also explore the question of whether this multiple-retrieval CBR could be an effective initialization method for local search methods such as hill climbing, tabu search and simulated annealing. Significant results are obtained from a wide range of experiments. An evaluation of the CBR system is presented and the impact of the approach on timetabling research is discussed. We see that the approach does indeed represent an effective initialization method for these approaches.     

Modified nonmonotone Armijo line search for descent method

Nonmonotone line search approach is a new technique for solving optimization problems. It relaxes the line search range and finds a larger step-size at each iteration, so as to possibly avoid local minimizer and run away from narrow curved valley. It is helpful to find the global minimizer of optimization problems. In this paper we develop a new modification of matrix-free nonmonotone Armijo line search and analyze the global convergence and convergence rate of the resulting method. We also address several approaches to estimate the Lipschitz constant of the gradient of objective functions that would be used in line search algorithms. Numerical results show that this new modification of Armijo line search is efficient for solving large scale unconstrained optimization problems.     

Computing genus 2 curves from invariants on the Hilbert moduli space

We give a new method for generating genus 2 curves over a finite field with a given number of points on the Jacobian of the curve. We define two new invariants for genus 2 curves as values of modular functions on the Hilbert moduli space and show how to compute them. We relate them to the usual three Igusa invariants on the Siegel moduli space and give an algorithm to construct curves using these new invariants. Our approach simplifies the complex analytic method for computing genus 2 curves for cryptography and reduces the amount of computation required.     

A simple finite volume method for the shallow water equations

We present a new finite volume method for the numerical solution of shallow water equations for either flat or non-flat topography. The method is simple, accurate and avoids the solution of Riemann problems during the time integration process. The proposed approach consists of a predictor stage and a corrector stage. The predictor stage uses the method of characteristics to reconstruct the numerical fluxes, whereas the corrector stage recovers the conservation equations. The proposed finite volume method is well balanced, conservative, non-oscillatory and suitable for shallow water equations for which Riemann problems are difficult to solve. The proposed finite volume method is verified against several benchmark tests and shows good agreement with analytical solutions.     

Recovering Beam Search: Enhancing the Beam Search Approach for Combinatorial Optimization Problems

A hybrid heuristic method for combinatorial optimization problems is proposed that combines different classical techniques such as tree search procedures, bounding schemes and local search. The proposed method enhances the classic beam search approach by applying to each partial solution corresponding to a node selected by the beam, a further test that checks whether the current partial solution is dominated by another partial solution at the same level of the search tree. If this is the case, the latter solution becomes the new current partial solution. This step allows to partially recover from previous wrong decisions of the beam search procedure and can be seen as a local search step on the partial solution. We present here the application to two well known combinatorial optimization problems: the two-machine total completion time flow shop scheduling problem and the uncapacitated p-median location problem. In both cases the method strongly improves the performances with respect to the basic beam search approach and is competitive with the state of the art heuristics.     

Numerical solution of the fully non-linear weakly dispersive serre equations for steep gradient flows

We demonstrate a numerical approach for solving the one-dimensional non-linear weakly dispersive Serre equations. By introducing a new conserved quantity the Serre equations can be written in conservation law form, where the velocity is recovered from the conserved quantities at each time step by solving an auxiliary elliptic equation. Numerical techniques for solving equations in conservative law form can then be applied to solve the Serre equations. We demonstrate how this is achieved. The system of conservation equations are solved using the finite volume method and the associated elliptic equation for the velocity is solved using a finite difference method. This robust approach allows us to accurately solve problems with steep gradients in the flow, such as those generated by discontinuities in the initial conditions.The method is shown to be accurate, simple to implement and stable for a range of problems including flows with steep gradients and variable bathymetry.     

All roads lead to Rome—New search methods for the optimal triangulation problem

To perform efficient inference in Bayesian networks by means of a Junction Tree method, the network graph needs to be triangulated. The quality of this triangulation largely determines the efficiency of the subsequent inference, but the triangulation problem is unfortunately NP-hard. It is common for existing methods to use the treewidth criterion for optimality of a triangulation. However, this criterion may lead to a somewhat harder inference problem than the total table size criterion. We therefore investigate new methods for depth-first search and best-first search for finding optimal total table size triangulations. The search methods are made faster by efficient dynamic maintenance of the cliques of a graph. This problem was investigated by Stix, and in this paper we derive a new simple method based on the Bron-Kerbosch algorithm that compares favourably to Stix’ approach. The new approach is generic in the sense that it can be used with other algorithms than just Bron-Kerbosch. The algorithms for finding optimal triangulations are mainly supposed to be off-line methods, but they may form the basis for efficient any-time heuristics. Furthermore, the methods make it possible to evaluate the quality of heuristics precisely and allow us to discover parts of the search space that are most important to direct randomized sampling to.     

Numerical simulations of water–gas flow in heterogeneous porous media with discontinuous capillary pressures by the concept of global pressure

We present an approach and numerical results for a new formulation modeling immiscible compressible two-phase flow in heterogeneous porous media with discontinuous capillary pressures. The main feature of this model is the introduction of a new global pressure, and it is fully equivalent to the original equations. The resulting equations are written in a fractional flow formulation and lead to a coupled degenerate system which consists of a nonlinear parabolic (the global pressure) equation and a nonlinear diffusion–convection one (the saturation equation) with nonlinear transmission conditions at the interfaces that separate different media. The resulting system is discretized using a vertex-centred finite volume method combined with pressure and flux interface conditions for the treatment of heterogeneities. An implicit Euler approach is used for time discretization. A Godunov-type method is used to treat the convection terms, and the diffusion terms are discretized by piecewise linear conforming finite elements. We present numerical simulations for three one-dimensional benchmark tests to demonstrate the ability of the method to approximate solutions of water–gas equations efficiently and accurately in nuclear underground waste disposal situations.     

An effective greedy heuristic for the Social Golfer Problem

The Social Golfer Problem (SGP) is a combinatorial optimization problem that exhibits a lot of symmetry and has recently attracted significant attention. In this paper, we present a new greedy heuristic for the SGP, based on the intuitive concept of freedom among players. We use this heuristic in a complete backtracking search, and match the best current results of constraint solvers for several SGP instances with a much simpler method. We then use the main idea of the heuristic to construct initial configurations for a metaheuristic approach, and show that this significantly improves results obtained by local search alone. In particular, our method is the first metaheuristic technique that can solve the original problem instance optimally. We show that our approach is also highly competitive with other metaheuristic and constraint-based methods on many other benchmark instances from the literature.     

一种新的无约束优化线搜索算法

在对各种有效的线搜索算法分析的基础上,给出了一种求解光滑无约束优化问题的新的线搜索算法.对于目标函数是二次连续可微且下有界的无约束优化问题,算法具有与Wolfe-Powell线搜索算法相同的理论性质.在每一步迭代中算法至多需要计算两次梯度,对于计算目标函数梯度花费较大的情形可以节省一定的计算量.数值试验表明本文算法是可行的和有效的.     

Non-overlapping Domain Decomposition Method for a Nodal Finite Element Method

A new approach is proposed for constructing nonoverlapping domain decomposition procedures for solving a linear system related to a nodal finite element method. It applies to problems involving either positive semi-definite or complex indefinite local matrices. The main feature of the method is to preserve the continuity requirements on the unknowns and the finite element equations at the nodes shared by more than two subdomains and to suitably augment the local matrices. We prove that the corresponding algorithm can be seen as a converging iterative method for solving the finite element system and that it cannot break down. Each iteration is obtained by solving uncoupled local finite element systems posed in each subdomain and, in contrast to a strict domain decomposition method, is completed by solving a linear system whose unknowns are the degrees of freedom attached to the above special nodes.     

Time aggregated Markov decision processes via standard dynamic programming

This note addresses the time aggregation approach to ergodic finite state Markov decision processes with uncontrollable states. We propose the use of the time aggregation approach as an intermediate step toward constructing a transformed MDP whose state space is comprised solely of the controllable states. The proposed approach simplifies the iterative search for the optimal solution by eliminating the need to define an equivalent parametric function, and results in a problem that can be solved by simpler, standard MDP algorithms.     

Discrete artificial boundary conditions for nonlinear Schrödinger equations

In this work we construct and analyze discrete artificial boundary conditions (ABCs) for different finite difference schemes to solve nonlinear Schrödinger equations. These new discrete boundary conditions are motivated by the continuous ABCs recently obtained by the potential strategy of Szeftel. Since these new nonlinear ABCs are based on the discrete ABCs for the linear problem we first review the well-known results for the linear Schrödinger equation. We present our approach for a couple of finite difference schemes, including the Crank–Nicholson scheme, the Dùran–Sanz-Serna scheme, the DuFort–Frankel method and several split-step (fractional-step) methods such as the Lie splitting, the Strang splitting and the relaxation scheme of Besse. Finally, several numerical tests illustrate the accuracy and stability of our new discrete approach for the considered finite difference schemes.     

Adaptive error control during gradient search for an elliptic optimization problem

In this article we describe a cost effective adaptive procedure for optimization of a quantity of interest of a solution of an elliptic problem with respect to parameters in the data, using a gradient search approach. The numerical error in both the quantity of interest and the computed gradient may affect the progression of the search algorithm, while the errors generally change at each step during the search algorithm. We address this by using an accurate a posteriori estimate for the error in a quantity of interest that indicates the effect of error on the computed gradient and so provides a measure for how to refine the discretization as the search proceeds. Specifically, we devise an adaptive algorithm to refine and unrefine the finite element mesh at each step in the search algorithm. We give basic examples and apply this technique to a model of a healing wound.