Aleksandr Katrutsa

1. Heavy-ball 2. Theory 3. Nesterov acceleration

Add in presentation compare the Frank-Wolfe and Projected gradient methods in case of problem with matrix domain

Take GeomD from [this paper](https://arxiv.org/pdf/1506.08187.pdf) and its composite option from [this paper](http://ai.tencent.com/ailab/media/publications/Geometric_Descent_Method_for_Convex_Composite_Minimization-_wei_liu.pdf)

Adopt algorithms from this [lib](https://www.mathworks.com/matlabcentral/fileexchange/20504-submodular-function-optimization) and use [tutorial](https://www.inf.ethz.ch/personal/ladickyl/cvpr_tutorial_part2.pdf), [review](https://hal.archives-ouvertes.fr/hal-00645271/document) and resources from [this site](https://las.inf.ethz.ch/submodularity/)

Add implementations of main algorithms from the [book](https://www.dropbox.com/s/6wd93z98m49ryrx/prox_algs.pdf?dl=0) and mini-tutorial in examples for usage of such algorithms, its interpretations, pros and cons.

1. Compute norms 2. Compute matrix factorizations 3. Matrix derivatives 4. Solve linear system 5. Compute eigendecomposition and others