[mlpack] mlpack:GSOC'17

[Ryan Curtin]

On Fri, Feb 10, 2017 at 12:16:02AM +0530, abhinav kumar wrote:
> Hello everyone,
>  Myself Abhinav Kumar, 4th year computer science student from NIT Srinagar,
> India. I am a machine learning and parallel computing enthusiast and worked
> on unsupervised learning during my previous internship. I was learning
> about ml programing then I came through mlpack. Its a nice and promising
> software.
> I have experience of using armadillo,cmake and lapack for my ml and c++
> programs. I have made contribution to gnome software.
> I like to contribute to mlpack and be part of GSOC'17 through mlpack
> organisation. I have compiled mlpack from source on my computer and used
> some of its algorithm. I am also familiar with git system and have a good
> experience with c++. I have read about template SFINAE and reading about
> policy based design (as provided in mlpack documentation).
> 
> I was going through mlpack GSOC list 2017 and I like to work on -
>  * Parallel stochastic optimization methods *
> I am very much interested to work on this project.I have found these paper
> related to this project -
>  for SCD (https://arxiv.org/pdf/1311.1873.pdf)
>  for SGD (http://martin.zinkevich.org/publications/nips2010.pdf)
> I have also found that this project was listed in GSOC 2015 and archive
> related to this is - Archive March 2015
> <http://knife.lugatgt.org/pipermail/mlpack/2015-March/001658.html>
>  For Martin Zinkevich paper, it was commented as
> " I think it makes more sense in a distributed setting (where communication
> is much more expensive than in a shared memory
> setting). " So, I have to look for other paper on it.
> For this project , I need to learn convex optimization and this course by
> Stanford University
> <http://online.stanford.edu/course/convex-optimization-winter-2014> can be
> really helpful. If there is any other resources to look for please direct
> me to that.
> I have some experience of creating multi-thread program on c# and i think
> by using that knowledge and studying tutorials based on c++ threads I can
> do a great work on this project.
> 
> It would be a great opportunity to be part of this organisation and
> contribute to mlpack.

Hi Abhinav,

There are several other mailing list threads, but since you have already
searched the archive, you have probably found them.  An implementation
should be written with OpenMP to match the rest of the parallel code in
mlpack, so keep that in mind while you are preparing your plans.

Let me know if there is anything else I can clarify.

Thanks,

Ryan

-- 
Ryan Curtin    | "And tell the engineers to wipe that stupid smile
ryan at ratml.org | off his face this time."  - Lou