You will be working on a statistical learning (estimation) method that we have recently developed in our group. In this method, the unknown parameters of a statistical model are estimated by solving an unconstrained optimization problem. We are solving the optimization problem iteratively and use currently at each update step all the data. We think, however, that the optimization can be sped up by using only small parts of the data in each update step.  In your work, you will be investigating whether known optimization techniques for high-dimensional problems can be used in that way to speed up the learning without affecting the accuracy of the estimate.

Requirements:
-Ability to program in MATLAB
-Knowledge of math, machine learning and statistics as required in our lecture course "Unsupervised Machine Learning"
-Ability to write reports in latex
- Some prior knowledge of optimization theory would be a plus

Type of work:
50% Theory (optimization methods and estimation)
50% Programming and simulations (implementation and testing of the methods)