In this thesis, we slightly modify a known routing model in packet-switching networks. Those modifications give priority to messages transiting in the network (internal) over external arrivals (users) incoming to each node while reducing the number of lost packets. Mathematical programs are developed for the multiple and single buffer node models. Those take into account line capacities, finite buffers and channel errors during transmission. We investigate a simple decentralized routing policy, based on the proposed centralized model with multiple buffers, where partial information is available from neighboring nodes only. Integer linear programming is used to optimally solve the routing problem for each of the models studied. Simulation, conducted on a 5-node network, illustrates the impact of these modifications.

This thesis presents a bioeconomic model of a commercial fishery. Emphasis is on the decision making processes of the harvesting sector. A model of dynamic decision making by fishermen is developed to study the biological and economic impact of the commercial exploitation of the Georges Bank scallops fishery (Placopecten Magellanicus). Stock biomass abundance dynamics are modelled using Deriso's age structured population model. Growth rate, mortality rates and recruitment are included as part of the biological component of the model. Decision making by fishermen is modelled using two discrete decision algorithms, myopic and adaptive. The myopic model provides the fisherman with the area yielding the highest immediate expected return on the basis of cost, expected catch and the current measure of biomass abundance. A computer model simulates the application of both the myopic and adaptive algorithms by a single fisherman as well as by a fleet of fishermen. In the later case, each fisherman is assumed to be initially identical and independent of the others in the fleet with respect to landings and catch information compiled. Vessel performance is measured in terms of total catch, total costs, landed values and net incomes. (Abstract shortened by UMI.)

To optimize the function of a prototype flicker perimeter, it is important to reduce the amount of time required to change test target location in the visual field while maintaining the random presentation order necessary to ensure accurate results. The purpose of this project is to develop and test a software algorithm that achieves these two goals simultaneously. I have developed an algorithm for the Flicker perimeter that maximizes the efficiency of travel in the motors governing test target projection. This will reduce the time between test trials as well as the entire sequence, thus limiting the fatigue factor in testing. The algorithm incorporates a component that maintains a pseudo-random order of presentation. I have developed and constructed a simulation of the functioning perimeter by designing and reengineering the system program and incorporating both the original presentation algorithm (Random) and the optimizing algorithm (Optimized). (Abstract shortened by UMI.)

We learn decision lists over a space of features that are constructed from the data. A practical machine which we call the Decision List Machine comes as a result. We construct the Decision List Machine which uses generalized balls as data-dependent features. We compare practical performance on some data sets with the performance of some other learning algorithms such as the Set Covering Machine and the Support Vector Machine. This performance is evaluated for both symmetric and asymmetric loss coefficients. We also provide a theoretical assessment of the performance of the DAM by computing upper bounds of the generalization error.

Abstract Not Available.

This thesis focuses on several aspects of data perturbation for Linear Programming. Classical questions of degeneracy and post-optimal analysis are given a unified presentation, in a view of new interior point methods of linear programming. The performance of these methods is compared to the simplex algorithm; interior point methods are shown to alleviate some difficulties of representation and solution of linear programs. An affine scaling algorithm is implemented in conjunction with a simple rounding heuristic to asses the benefit of interior point trajectories to provide approximate solutions of linear integer programming.

Program-understanding schemata for understanding the static properties of C++ programs are presented. For understanding purposes, we have selected seven key concepts of the C++ programming language. A C++ program is viewed as being composed of these seven key concepts and the relationships which may exist between them. These seven key C++ concepts and their relationships are represented using concept templates. The information grouped within these concepts is shown using twenty-six display templates. A representation schema and a navigation schema for the concept templates and the display templates are explained in detail.

This thesis explores applications of Interior Point methods as popularized by Karmarkar (36) for solving Multicommodity Network Flow problems (MCNF). In these problems, several commodities must be shipped between various nodes of a network. The goal is to satisfy the shipping requirements at a minimum cost, while respecting processing capacities on the joint flow of commodities. The thesis presents a unified view of current methods for multicommodity networks, from the early formulations to current work that has involved nonlinear and interior point methods. It compares and contrasts some Interior Point methods, and discusses how they can be applied to the MCNF problem. It builds upon the first Interior Point partitioning method proposed to solve MCNF, and implements a related partitioning method. This scheme allows network subproblems to be solved using efficient network simplex algorithms: then the results are coordinated using a linear program that is solved with an Interior Point affine scaling algorithm. It presents computational results, comparing the results from the partitioning method with those found by solving an LP formulation of the problem using the affine scaling method.

Stability and efficiency (i.e. derivative function evaluations per processor) are the two main considerations in deriving good numerical methods for ODE's. The underlying challenge is to increase the stability region while maintaining or even improving efficiency. To achieve this, some extensions of predictor-corrector based methods, which apply a fixed number of corrector iterations, are considered. This thesis studies two particular members of a family of methods called the Parallel Block Predictor-Corrector Family, which are based on these extensions. These two members are called PBPC/2 and PBPC/3. They are characterized by iterated corrector evaluations carried out in two adjacent blocks. Stability properties of these methods are analyzed and compared with some existing block-based parallel predictor-corrector methods. Performance of the PBPC/2 and PBPC/3 methods and these existing block-based parallel predictor-corrector methods is compared using solution formulas which extend over a range of integration orders and which use various number of processors. The results obtained from a stability analysis and from a collection of numerical experiments indicate that the proposed methods provide a potential opportunity to balance stability properties and efficiency in the parallel computer systems.

This thesis evaluates the benefit of meshing mathematical programming and expert systems for solving capital budgeting problems, using constraint logic programming methods. A review of modelling capabilities of mathematical programs for capital budgeting, and of financial expert systems leads to defining the respective role and potential of each method, and to the proposal of a two-tiered project selection approach: project evaluation and resource allocation. With emphasis placed on a tight coupling of the two tiers, logic programming is shown to be a language of choice to implement mathematical programming within an expert system shell. Prolog has the requisite properties to deal with both logical considerations and optimization problems. Although Prolog was not primarily designed to solve optimization problems, it is shown that the backtracking mechanism of the Prolog language is powerful enough for that purpose; it liberates the programmer from having to implement tree-search programs. A generate and test program is written in Turbo-Prolog, and compared to a more sophisticated test and generate implementation that uses methods of constraint satisfaction programming. Continuous capital budgeting problems are solved in CLP(${\cal R}$), an experimental extension of Prolog that enables the solution of simultaneous algebraic constraints, as required to solve linear programs.