Information Processing

This project is split into two parts. The first is to model the arrival process to take into account the forms of dependence in the forecast call center of Hydro-Quebec. A second aspect concerns the proposal for a control system for real-time reallocation of agents. For the first part the need for more accurate models of stochastic processes has been made by officials of Hydro-Quebec. We therefore propose a model incorporating both dependence and intraday dependence between call types. It would then provide better predictions of future arrival of calls during the day.

Business Intelligence (BI) refers to computer-based techniques that facilitate the use of information within organizations to make informed decisions and to run operations effectively based on available data. Our goal is the research and development of new BI tools for focused search and retrieval, which will support analysis and decision-making related to energy conservation and management. Such tools will also have wide applicability to business analytics in general.

The research is aimed at providing effective long-term resource planning to effective scheduling of the maintenance tasks over a short-term horizon. The Bombardier Company provides the necessary requirements to the customers around the world to do the predefined maintenance tasks as well as unexpected repair jobs for their aircraft fleet. These services are performed as onsite or offsite, i.e., different centres or stations.

Novel methods for the computationally efficient simulation of compressible fluid flow will be developed, with applications to aerospace and space physics. First, new formulations of multigrid techniques for the implicit solution of time-dependent flows will be studied, using parabolization to increase diagonal dominance and solution efficiency. Second, new direct solution methods for steady transonic flows will be developed, employing dynamical systems and characteristic analysis.

Modern engineering designs require fast and high credible scientific computations which usually run in a parallel way. The proposed research focuses on the development of the parallel preconditioning technology used in large scale scientific computations. A multi-level recursive strategy is developed to improve the parallel computing performance when a large number of processors (up to at least 5000) are used. An existing Newton-Krylov flow solver will be improved by coupling with this multi-level preconditioner.

The theoretical and practical aspects of manipulating mathematical expressions on computers are usually referred to as computer algebra or symbolic computation. In this field, calculations are designed to yield exact and complete results, by opposition to numerical analysis which is meant to handle approximate values, potentially producing incomplete results. Exactness and completeness have some significant computational overhead. Computer algebra software is highly demanding in CPU time and memory.

We will develop algorithms to automatically generate descriptive labels for large collections of web documents. Such labels can be used by companies in order to decide on which web sites they want to place advertisements, or by electronic publishers to categorize media offers. Currently, there doesn't exist any approach that can robustly and automatically label clusters of documents with a level of quality that approaches human labellers.

This project uses mathematical filtering theory to develop computer tractable real time solutions for incomplete, corrupted information problems. These techniques have proven to be beneficial in defence, communications, media effects, and manufacturing. In 2002-2003, Optovation Inc. was added as a new partner, Lockheed Martin Corp. filed for two new patents and we formed a spin-off company, Random Knowledge Inc. to commercialize our technology in the areas of Network Security, Fraud Detection, and Finance.