CSE Profile: Daniel Boley

Daniel Boley
Professor

(612) 625-3887
Office: EE/CS 6-209

Interests

Numerical analysis, control theory, linear algebra, iterative methods, machine learning, and robotics.

Education

Ph.D. 1981, MS 1976, Computer Science, Stanford University

B.A. 1974, Mathematics, Cornell University

Professor Boley specializes in numerical analysis, linear algebra and control, computational methods in statistical machine learning and unsupervised document categorization in data mining and bioinformatics.

He is an IEEE senior member and a fellow in the Minnesota Supercomputer Institute. Boley has approximately 140 publications and has served on numerous professional panels and committees. He is a member of the Society for Industrial and Applied Mathematics, ACM, an ACM special interest group on Numerical Mathematics, and the IEEE Computer Society. Boley is also an associate editor for the SIAM Journal of Matrix Analysis and has chaired several technical symposia at major conferences.

In addition to his work as a professor, Boley has had extended visiting positions at the Los Alamos Scientific Laboratory, the IBM Research Center in Zurich (Switzerland), the Australian National University in Canberra, Stanford University, and the University of Salerno (Italy).

Research

My research interests include large sparse linear algebra problems arising from many engineering applications, the integration of numerical techniques in new technologies in a robust and fault tolerant manner, and the application of similar techniques to specific applications. Applications include control theory, electromagnetics, robotics, unsupervised machine learning, and vehicle navigation.

In machine learning, techniques related to spectral graph partitioning lead to unusually fast and effective methods for unsupervised clustering. This is an example of how fast linear algebra methods find their way into new areas. Using these techniques, we are exploring very large datasets derived from text documents, textile images, movie ratings, voice transcription data, etc. They also serve as a basis for a client-side Web agent capable of automatically organizing documents retrieved through the browser for the user.

In control theory, identification of systems from signal data using efficient state-space linear algebra techniques is an ongoing research effort. Our goal is to improve the robustness and numerical stability of the methods while not losing the advantages of a fast algorithm. Similar techniques carry over to the design and synthesis of controllers.

In electromagnetics and many similar applications in physics and engineering, the resonant modes lead naturally to very large sparse matrix eigenvalues, for which the efficient solution is sought. Iterative techniques lead to efficient solutions, but these depend on an appropriate ordering of the unknowns for sparsity, serial time complexity, and parallel efficiency.

Personal Home Page