Friday, April 30th, 2004 at 11:00a.m.

Human face recognition is one of the classical problems in computer vision with many potential applications such as searching large face databases, human computer interface, video surveillance and biometric security. One of the most widely accepted and promising approaches in face recognition is Principal Component Analysis (PCA). Although PCA has its limitations: large computational load and poor adaptability with change in environment. In this thesis we evaluate the use of Random Projection for face recognition. Random Projections (RP) has been identified as a promising approach for dimensionality reduction. RP is significantly less expensive compared to other conventional techniques. It is this property of RP that makes it a preferred approach over PCA in face recognition algorithm. Our results are based on varying projecting dimensions and different random ensembles. Over the experiments we compare our results with the Eigen face approach.

Wednesday, December 10th, 2003 at 3:00p.m.

This thesis describes work done in optimizing an existing NeoCortical Simulation Program (NCS), including the development of a set of parallel profiling and measurement tools.

The NCS program is an ongoing project of the Brain Computation Lab. Previous development work was most recently presented in E. Courtenay Wilson's thesis. It will be shown that from that base, this work has increased sequential computation speed by at least an order of magnitude; increased the demonstrated model size by three orders of magnitude; created a program which exhibits near-linear speedup over the number of processors tested; and, despite having added significant additional functionality, has decreased the code base by some 45 percent.

Monday, December 8th, 2003 at Noon

Text classification is a problem that grows more relevant each day. It was recently estimated that there are almost 800 MB of data produced per person, per year, with 92% of that data being stored on magnetic media. Much of this data is text (memos, reports, emails, Web pages) which is often classified according to some criteria set by the user. The ability to automatically classify much of this data would save time and improve the useability of computers systems. In this thesis, we demonstrate a technique for improving the performance of an automatic text classification system in a hierarchical class space as well as a flat class space.

Friday, December 5th, 2003 at 4:00p.m.

Large groups of networked workstations, commonly referred to as Beowulf clusters, require a systematic approach to load balancing. Many applications require extensive message passing and synchronization to take full advantage of the available processing power. We have endeavored to simplify this task by developing a generic queuing system that can be adapted to different applications. This system is particularly suited to graph theory problems, many of which require a high ratio of computation to message passing. We have use the queuing system to solve a common graph theory problem, finding the Minimum Crossing Number of a complete graph.

Friday, December 5th, 2003 at 3:00p.m.

Combining formal and semi-formal methods to create software specifications capitalizes on the advantages of both styles of notation. Semi-formal notations such as UML allow the creation of easy to understand descriptions of the systems architecture and behavior. Formal notations such as Z++ allow precise elaboration of specifications that can be refined up to the point of implementation. Previously proposed translation rules for the conversion of UML statecharts to Z++ specifications have not included support for nested statecharts. Nested statecharts, on which the approach presented in this paper is focused, are powerful modeling constructs and thus hold strong importance in correctly developing complex models of software systems. Because user interfaces can be mission critical in certain types of application, the combination of formal and semi-formal notations for user interface specification is explored in this paper. In addition, rules for translating UML statecharts to Z++ are presented and the prototype of a Z++/UML editor is described.