Optimization for Statistical Modeling. 3 credits

Catalog description

Discusses standard classes of mathematical optimization problems and how these classes arise in statistical model fitting. Both constrained and unconstrained optimization problems are studied in detail, with an emphasis on convex problems. Specific examples are: sparsity and shape-constrained estimation, EM algorithms for mixture models, linear programming for quantile regression, semidefinite programming for sparse PCA and Gaussian graphical models. The treatment is complemented by the implementation of suitable algorithms for the solution of the above problems, including gradient descent and proximal methods, Newton and Quasi-Newton methods, interior point methods, alternating direction methods of multipliers, and MM algorithms. Offered by Statistics . May not be repeated for credit.