Research Projects and Publications

Projects

• Randomized Fast Subspace Descent Scheme
  1. Developed a general scheme of gradient methods.
  2. Analyzed convergence for both convex and strongly convex problems.
  3. Numerical experiments on Nestrov’s worst function and Ridge Penalized Logistic Regression.
• Non-uniform Sampling Fast Least Squares Solver
  1. Developed fast solver for large scale ill-conditioned least squares problem.
  2. Constructed preconditioner via importance sampling and Gauss Aeidel Algorithms.
  3. Improved Performance up to 70% compared to diagonal precoditioned CG.
  4. Handled largest matrix size around 27 million.

Preprints

A Preconditioner Based on Non-Uniform Row Sampling for Linear Least Sqaures Problems