[mlpack] GSoC 2017: Reinforcement Learning and Essential Deep Learning Modules

Abhinav Moudgil abhinavmoudgil95 at gmail.com
Fri Mar 24 14:00:40 EDT 2017


Hi Marcus,

I am glad to hear back from you. Taking a detailed look at both the
projects, I have decided to go ahead with the second project i.e.
"Essential Deep Learning Modules". Just to reiterate, I am planning to
implement GAN, BRN and RBM. I will draft a proposal and share it with you
on Google Summer of Code website for feedback.

Kind regards,
Abhinav Moudgil

On Fri, Mar 24, 2017 at 10:25 PM, Marcus Edel <marcus.edel at fu-berlin.de>
wrote:

> Hello Abhinav,
>
> thanks for getting in touch and welcome.
>
> I would love to contribute to mlpack during this summer. It would be great
> if
> you could elaborate your views on the above projects. Looking forward to
> your
> guidance.
>
>
> That are some really neat project's you listed above. I think Ryan tried
> something "similar" as training a language model on a jokes corpus. Anyway
> here
> are my two cents to the projects mentioned above, I think each project is
> equally interesting and depending on what you like to do equally difficult
> and
> at the same time rewarding. The intention behind each project is to work on
> recent ideas and to provide a fast implementation at the end of the
> summer. At
> the end I can't help you with the decision since you worked on each topic
> it's
> even difficult to give you a recommendation.
>
> I hope something I said was helpful,
>
> Thanks,
> Marcus
>
> On 23 Mar 2017, at 16:25, Abhinav Moudgil <abhinavmoudgil95 at gmail.com>
> wrote:
>
> Hi,
>
> I am Abhinav Moudgil, a senior undergraduate research student in Deep
> Learning and Computer Vision, working on PR #942
> <https://github.com/mlpack/mlpack/pull/942>. I went through mlpack
> project ideas
> <https://github.com/mlpack/mlpack/wiki/SummerOfCodeIdeas#essential-deep-learning-modules>
> and I found the following two projects really interesting (in preference
> order) for Google Summer of Code 2017:
>
> *1. Reinforcement Learning (RL)*
> It would be a great learning experience for me to implement RL algorithms,
> which are fast and scalable. Previously, I have studied various RL
> algorithms well like Monte Carlo Policy Gradient (PG) with REINFORCE
> <https://gist.github.com/abhinavmoudgil95/138db4c55c42f91f4c858294acadb771>,
> Deep Q-learning (for discrete and continuous state space), Deep
> Deterministic PG with Actor-Critic networks, Policy Iteration for Maze
> environment, Hill Climbing
> <https://gist.github.com/abhinavmoudgil95/108123c880488965b8c1744cacd60dd6>,
> Random Search
> <https://gist.github.com/abhinavmoudgil95/6fcb2db7314e6c4f6b7a028dfe1f27db> etc.
> I have implemented and tested them in Python using Tensorflow. My OpenAI
> gym profile is accessible here
> <https://gym.openai.com/users/abhinavmoudgil95>. I will open source all
> my RL codes in a separate repository soon.
>
> *2. Essential Deep Learning Modules*
> I have studied the relevant literature for this project in the past and I
> like converting mathematical equations from research papers to code. In
> Summer 2016, I worked on feature engineering as a Google Summer of Code
> project <https://abhinavmoudgil95.github.io/2016-08-23/gsoc-conclusion/>
> with CERN SFT where I worked on some advanced feature extraction methods
> like Deep Autoencoders, Feature Clustering, Hessian Locally Linear
> Embedding etc. So, I explored literature on Restricted Boltzmann Machines,
> Hopfield Networks etc. In this project, I would like to implement the
> following models:
>
>    - RBM - Studied extensively during my Google Summer of Code, 2016.
>    - GAN - This semester, I am a Teaching Assistant for the course
>    Statistical Methods in AI at my university IIIT-H
>    <https://www.iiit.ac.in/>. As a part of this job, I am mentoring
>    projects like Coupled GANs, Conditional GANs. I have studied the GAN
>    literature well along with its variations like DCGANs, Improved Techniques
>    for training GANs by OpenAI, Class Conditional GANs by Yann Lecun etc.
>    - BRN - I solved
>    <https://abhinavmoudgil95.github.io/2017-03-01/funnybot/> OpenAI
>    Request for Research problem #2
>    <https://openai.com/requests-for-research/#funnybot>. For that, I
>    studied Recurrent Neural Networks in detail along with variations of it
>    like LSTMs, Attention Models, BRNs. Currently, I am working on OpenAI
>    Request for Research #3
>    <https://openai.com/requests-for-research/#im2latex> which involves
>    implementing Attention Models and Bidirectional RNNs.
>
> *Open Source Experience: *
> I worked <https://github.com/abhinavmoudgil95/gsoc-2016> with CERN SFT on
> feature engineering module as a Google Summer of Code student. I
> contributed
> <http://wiki.opencog.org/wikihome/index.php/Special:Contributions/Amod95>
> to OpenCog foundation by fixing several bugs and writing an installation
> script <https://github.com/opencog/ocpkg/pull/50> for Mac OS X. I also
> contributed to Shogun, a Machine Learning toolbox where I worked on
> improving and benchmarking
> <https://github.com/shogun-toolbox/shogun/issues/3048> basic ML
> algorithms like PCA, LDA etc.
>
> I would love to contribute to mlpack during this summer. It would be great
> if you could elaborate your views on the above projects. Looking forward to
> your guidance.
>
> Kind regards,
>
> Abhinav Moudgil
> Github: https://github.com/abhinavmoudgil95
> Website: https://abhinavmoudgil95.github.io/
> LinkedIn: https://www.linkedin.com/in/abhinavmoudgil/
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20170324/9ee7d48d/attachment-0001.html>


More information about the mlpack mailing list