Anthony Nwachukwu
supervisor: Andrzej Karbowski
We considered several algorithms used to solve separable, nonconvex problems and implemented five of them, Bertsekas, Tanikawa-Mukai, Tatjewski, Arnold, et al. and ADMM using MATLAB. Experiments were performed using the implementations on a machine learning convex problem, the ridge regression for model fitting using an artificially crafted dataset. From the experiments, we observed that although the work of Tanikawa-Makai obtained convergence in a slightly smaller iteration than the rest, it takes a longer time per iterations compared to some others. Also, ADMM and Bertsekas achieved convergence at a close number of iterations, while Tanikawa-Mukai has the advantage of shorter run-time. We concluded that Bertsekas gave the best performance given the problem considered while Arnold and Tatjewski gave the worst performance.