[mlpack] GSOC 2019

[Marcus Edel]

Hello Bhavya,

thanks for getting in touch.

> I am mainly interested in two projects: 1) Essential deep learning modules: I
> was thinking of some possible projects like highway networks and bidirectional
> attention flow for machine comprehension but I am not sure if it would be
> feasible to train it on a question answering dataset without a GPU. Please let
> me what you think!

Good point, I'm not sure if building against NVBLAS would be helpful here, and
Bandicoot (GPU accelerator for Armadillo) is still in an early stage. However,
if you still like the challenge I think Highway Networks is still a nice idea
that could be useful for other tasks as well.

> 2) mlpack-Tensorflow translator: I find the idea of translating models trained
> using one framework to another one very interesting as it would allow
> collaboration between people who are not familiar with different frameworks. I
> was thinking of building upon the work that has been done on CLI for ANN to get
> started with this one.

Agreed, the CLI idea is definitely a nice starting point, the model
serialization might be helpful as well; see http://mlpack.org/docs/mlpack-3.0.4/
doxygen/anntutorial.html#model_saving_loading_anntut for some additional
information.

Let me know if I should clarify anything else.

Thanks,
Marcus

> On 5. Mar 2019, at 05:41, Bhavya Bahl <bhavyabahl.1 at gmail.com> wrote:
> 
> Hello everyone,
> 
> My name is Bhavya Bahl and I am a computer science undergraduate at Indian Institute of Technology, Bombay. I have implemented some basic tests for local coordinate coding and have been going through mlpack codebase over the past one month. I am looking forward to spending my summers contributing to mlpack.
> 
> I am mainly interested in two projects:
> 1) Essential deep learning modules: I was thinking of some possible projects like highway networks and bidirectional attention flow for machine comprehension but I am not sure if it would be feasible to train it on a question answering dataset without a GPU. Please let me what you think!
> 2) mlpack-Tensorflow translator: I find the idea of translating models trained using one framework to another one very interesting as it would allow collaboration between people who are not familiar with different frameworks. I was thinking of building upon the work that has been done on CLI for ANN to get started with this one.
> 
> Any suggestions are welcome.
> 
> Regards 
> Bhavya
> _______________________________________________
> mlpack mailing list
> mlpack at lists.mlpack.org
> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack