8:30 AM-9:15 AM
Chair: Gene H. Golub, Stanford University
Kresge Auditorium
The least squares approximation of a given matrix by a rank deficient one can be obtained from its singular value decomposition (SVD). However, in general, this procedure does not preserve the structure, if any, in the data matrix (such as e.g. Hankel, Toeplitz, etc..).
Yet the combination of structure and rank deficiency is crucial in many applications, ranging from biomedical signal processing, over telecommunications, to system identification for model-based process control.
In this presentation, the speaker will discuss a nonlinear generalization of the SVD, called the Riemannian SVD, which delivers a rank deficient least squares approximation of a given data matrix, while preserving its (affine) structure. He will discuss geometrical, algebraic, and statistical properties and interpretations and numerical issues. Several practical engineering examples will be illustrated.
Bart De Moor
ESAT/SISTA, Katholieke Universiteit Leuven, Belgium
AN97 Homepage | Program Updates|
Registration | Hotel and Dormitory Information | Transportation | Program-at-a-Glance | Program Overview