[mlpack] Variational Autoencoders - GSOC project idea

[Ryan Curtin]

On Sat, Feb 24, 2018 at 07:17:19PM +0530, Atharva Khandait wrote:
> Hi Sumedh,
> 
> I am Atharva Khandait (github: akhandait, irc: Atharva). I am currently a
> sophomore at IIT Bombay.
> Over the past couple weeks, I solved some issues in mlpack and currently
> working on some. It helped me a lot and have started getting familiar with
> the source code and the community. :)
> 
> I found the VAE project idea very interesting. I have previously worked
> with quite a number of neural networks, some of them being convolutional
> and LSTMs. I also worked for a month for a team at IIT Bombay which is
> making a driverless car. There, I worked with an open source implementation
> of the CycleGAN architecture and picked up some knowledge on generative
> models as well.
> 
> I would love some suggestions on how I should proceed specifically towards
> this project as the proposal period comes near. I have my exams for the
> next 4 days, after that I would be free almost all the time and would like
> to dive into this more.

Hi Atharva,

Thanks for getting in touch.  The VAE project is a new project this year
so there is not a long mailing list history to consult for this one.
Sumedh may have some clarification on the project itself, but I can help
provide some basic ideas for what would be useful---

When you submit your proposal, be sure that you have thought through
what the API for VAEs will be and what will be provided to the user.
It's very important that at the end of the summer, the project won't be
an "invisible" part of the mlpack codebase, but instead the VAEs should
be easily findable by a new user.  So I would advise that you put some
thought into that.

I hope this is helpful.

Thanks,

Ryan

-- 
Ryan Curtin    | "He takes such wonderful pictures with
ryan at ratml.org | his paws."  - Radio man