Stanford 50: State of the Art and Future Directions of Computational Mathematics and Numerical Computing


  • March 30, 2007
  • 4:50 pm - 5:15 pm

A statistician's debt to numerical analysts

Grace Wahba (University of Wisconsin-Madison)

Statisticians, including this one, owe a huge debt to numerical analysts. Where would we be without the Singular Value Decomposition, Spline Algorithms, Matrix Computations (Golub and Van Loan)?

After briefly noting my collaboration with Gene and Michael Heath on Generalized Cross Validation (1979), which laid the foundation for much later work, I will describe some more recent work of my own and collaborators that relies on mathematical programming and convex cone algorithms for the numerical solution of large problems. These include Regularized Kernel Estimation for data sets with dissimilarity data rather than attribute data, and the LASSO-Patternsearch algorithm for finding patterns of high order interactions in risk factor models with large and extremely large attribute vectors.

Stanford University Home Page