A Reductionist Approach to Process Discovery
Abstract
The primary objective of this research program is to develop and apply mathematical tools for the purposes of process discovery. Our focus is on empirically based methods applied to massive data sets in a high-dimensional setting. The emphasis of the program is on applications of direct interest to the Air Force. In particular, we have been investigating problems of interest to Air Force Research Laboratory, Materials & Manufacturing Directorate, Materials Process Design Branch at Wright Patterson Air Force base. Lately, we have intended the work to a problem of interest to the United States Forest Service. Recent work includes the application of subspace noise reduction methods and their connection to blind source separation. We have established a theoretical connection between the maximum noise fraction method and independent component analysis and demonstrated the relationship with examples. This methodology has proven useful as an integral component of the Whitney Reduction Network, developed by the grantees. Additionally, a new approach for designing support vector machines has been developed for the classification problem using a kernel based Fisher discriminant method. In addition we have developed other algorithms in terms of kernel functions using a kernel Grim-Schmidt algorithm These techniques have been applied to the materials science bonding problem.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 31, 2002
- Accession Number
- ADA414328
Entities
People
- Michael Kirby
Organizations
- Colorado State University