Title: Optimization of Sequencing and Threshold Levels of Detection Systems

Data generated from different analytical methods, x-ray detectors, gamma-ray detectors and other sensors are often relied upon to make critical decisions with regard to the nature of the containers and the appropriate response mechanism. The process of designing an efficient and accurate detection system must be deliberate and well thought-out. The threshold levels of sensors at the inspection stations might result in accepting undesired containers or subjecting "good" containers to unnecessary additional inspections. We investigated approaches for determining the optimum arrangements of sensors and their corresponding threshold levels while considering potential measurement errors, cost and other constraints. Efficient approaches for investigating inspection systems with a large number of sensors are also investigated. The theoretical contributions of this research are transferred via computational algorithms for practical applications.

Title: Determining Common Authorship Among Documents

Traditionally, author identification focuses on assigning a document to a single author with a high level of confidence. Using existing techniques and software as a starting point, we attempt to determine whether two documents share a common author without considering the identity of that author. Initial results show high accuracy using both raw feature vectors and existing author identification models. Analyses using these existing models appear to be particularly effective at identifying pairs of documents with a common author.

Title: Privacy-preserving distributed data mining of hybrid fragmented data

Two important directions for data mining research are distributed data mining and privacy-preserving data mining. Potentially useful data is often distributed among multiple parties and it is not always feasible to aggregate all the data in one site before it is mined. Privacy issues are one major barrier to aggregation and unrestricted sharing of data. For these reasons, it is useful to have efficient techniques for mining data in a distributed environment while preserving private information. Current research has focused on data which is horizontally or vertically fragmented. However, in many real-world situations, data comes from heterogeneous sources and is neither horizontally, nor vertically distributed. In this talk, we will present an overview of some privacy-preserving data mining techniques and we will discuss the challenges of privacy-preserving mining of hybrid fragmented data.

Title: Picking Teams of Warfighters: Beyond the Playground

The completion of any type of special mission requires selecting the best team for the job. Here, we examine the best way to simultaneously compose such teams from a limited population. In doing so, each team must meet certain competency and resiliency requirements, as dictated by the mission. Our aim is to create an efficient algorithm to assign people to teams such that these requirements are met, while minimizing the number of people assigned to teams.

Page last modified November 20, 2007.