[mlpack] GSoC 2018: Particle Swarm Optimization

[Marcus Edel]

Hello Chintan,

welcome and thanks for getting in touch.

> I would like to apply for the GSoC 2018 project on adding the Particle Swarm
> Optimizer for constrained and unconstrained problems. I have some prior
> experience working on PSO ([1], [2]) and would love to work on it yet again. The
> first link is the project where we train a basic Neural Net using PSO. It is
> still a work in progress. The second link is just a bare bones implementation of
> PSO.

Looks really interesting, the "Better initialization" point is definitely
something that might affect the overall training time in a positive way.

> I have followed the getting started links provided on the GSoC 2018 wiki page
> and have built mlpack, and intend to tinker with it and go through the source
> code once I'm done with the ongoing exams (by 10 March, 2018).

Sounds good, best of luck with your exams.

> Another interesting idea I would like to present is the use of gradient descent
> with PSO in a hybrid optimization approach ([3]: the repo belongs to my project
> mate). As of now it gives better training results, but neither of out approaches
> have been tested enough to actually comment on the matter yet.

That is an interesting idea, I guess "Multi-objective Particle Swarm
Optimization with Gradient Descent Search" by Li Ma and Babak Forouraghi could
be interesting here as well.

I hope anything I said was helpful; let me know if I should clarify anything.

Thanks,
Marcus

> On 8. Mar 2018, at 14:37, Chintan Soni <chintan.soni4 at gmail.com> wrote:
> 
> Hello everyone,
> 
> My name is Chintan Soni and I'm a 4th year CS student at Pune Institute of Computer Technology, Pune, India.
> 
> I would like to apply for the GSoC 2018 project on adding the Particle Swarm Optimizer for constrained and unconstrained problems.
> I have some prior experience working on PSO ([1], [2]) and would love to work on it yet again.
> The first link is the project where we train a basic Neural Net using PSO. It is still a work in progress. The second link is just a bare bones implementation of PSO.
> 
> I am also currently interning at Nvidia Pune (till ~10 May, 2018) from where I have experience working with somewhat complex C++ code and TMP, so understanding the mlpack codebase shouldn't be much of a problem.
> 
> I have followed the getting started links provided on the GSoC 2018 wiki page and have built mlpack, and intend to tinker with it and go through the source code once I'm done with the ongoing exams (by 10 March, 2018).
> 
> Another interesting idea I would like to present is the use of gradient descent with PSO in a hybrid optimization approach ([3]: the repo belongs to my project mate). As of now it gives better training results, but neither of out approaches have been tested enough to actually comment on the matter yet.
> 
> Regarding the constrained optimization problem, I've come across an idea of using PSO to solve Max-CSPs in the past (I cannot find a link to the paper as of now but I'll try looking it up). Is that a step in the right direction? Also, could you please provide references to other approaches for the same? (Especially if you have anything specific in mind.)
> 
> I will draft a proposal as soon as possible. Really looking forward to working on the project. Thanks in advance for your help.
> 
> Regards,
> Chintan Soni
> 
> Links used:
> [1] https://github.com/chintans111/ANNPSO <https://github.com/chintans111/ANNPSO>
> [2] https://github.com/chintans111/SPSO <https://github.com/chintans111/SPSO>
> [3] https://github.com/munagekar/nnpso <https://github.com/munagekar/nnpso>
> _______________________________________________
> mlpack mailing list
> mlpack at lists.mlpack.org
> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20180308/38d6b32c/attachment-0001.html>