[mlpack] GSoc '18 Project Idea 'Reinforcement Learning'

[Marcus Edel]

Hello Rishi,

thanks for getting in touch.

Implementing NEAT is definitely an interesting project, there is an open PR:
https://github.com/mlpack/mlpack/pull/752 which provides a basic implementation,
however, it's not finished yet and does only work on simple tasks. The problem
we encountered is that the population doesn't evolve after some iterations.
Anyway, it would be great to have and this could be a great opportunity to work
on it.

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

Thanks,
Marcus

> On 27. Feb 2018, at 19:20, Rishi Kaushik <rishi.kaushik1596 at gmail.com> wrote:
> 
> Hello everyone,
> My name is Rishi Kaushik and I am a Computer Science major at BITS, Hyderabad, India. I am a Machine Learning enthusiast and am well versed with C/C++, Java and Python. I have done a few basic projects implementing ML algorithms in C++ and Java and the list of projects mlpack has offered really excites me. 
> I am particularly interested in the idea of the project 'Reinforcement Learning'. What I wanted to ask was, can this idea of training a neural network to play games be extended to more than Reinforcement Learning algorithms? For the past few weeks I have been looking into the NEAT (NeuroEvolution of Augmenting Topologies) genetic algorithm and have seen implementations using NEAT playing classic games. I must admit I have no knowledge about the efficiency of NEAT vs. RL for this specific project but neuro evolution techniques have been shown to outperform RL methods on benchmark tasks.
> Also as the description says we can choose one or two different algorithms I think it'll be interesting to implement a RL based technique along with one NEAT based implementation and compare the benchmark results.
> I am very much interested to learn more about this and build this project with you, please let me know how I should proceed and if the variation I have offered seems plausible or not.
> Thank You. 
> _______________________________________________
> mlpack mailing list
> mlpack at lists.mlpack.org
> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack