Dinkelbach’s algorithm for fractional programming
I am a postdoctoral fellow at the University of California Berkeley in the Department of Statistics and a Principal Investigator at the International Computer Science Institute. I am working with Michael Mahoney. Before that, I completed a PhD in optimization in the School of Mathematics at the University of Edinburgh under the supervision of Prof. Jacek Gondzio. As of September 2018, I will be an assistant professor at the University of Waterloo’s Cheriton School of Computer Science.
PhD in Numerical Optimization, 2015
University of Edinburgh
MSc in Operational Research and Computational Optimization, 2010
University of Edinburgh
BSc in Management Science and Technology, 2009
Athens University of Economics and Business
Local Graph Clustering provides methods to find local clusters in a given graph without touching the whole graph. C++ implementation with Python interface.
Trillion. Instance generators for l1-regularized over- and underdetermined least squares. The generators are implemented in MATLAB, they are memoryless and computationally inexpensive. Hence, large-scale instances can be created.
pdNCG: primal-dual Newton Conjugate Gradients. A MATLAB implementation for the solution of unconstrained l1-regularized problems. For example, Machine Learning problems, such as l1-regularized least-squares and logistic regression, Compressed Sensing problems, such as l1-synthesis, l1-analysis and isotropic total-variation. The solver is memoryless, it requires only matrix-vector product operations, hence it is appropriate for large-scale instances.
MFIPMCS: Matrix-free Interior Point Method for Compressed Sensing. An interior point method implemented in MATLAB for the solution of real valued compressed sensing problems. The “matrix-free” implies that only matrix-vector products operations are allowed and the process is memoryless. The solver implements efficient preconditioning techniques for the fast solution of linear systems at every iteration.
FCD: Flexible Coordinate Descent. This software only reproduces the experiments in paper Flexible Coordinate Descent.
Data-Driven Discovery of Models (D3M). Project duration 2016-2020. In collaboration with M. Mahoney, Farbod Roosta-Khorasani and Alex Gittens.
About research and subjects of personal interest
Dinkelbach’s algorithm for fractional programming
Fractional programming generalizes flow-based models and algorithms for local graph clustering