[mlpack] GSoC'18: Essential Deep Learning Modules

[Ewelina Nowak]

Hello Marcus,

thank you for the suggestions!

At this time both architectures (RCNN and R-CNN) seem interesting to me.
For my first proposition - RCNN I have knowledge and experience in
implementing this architecture, so it should be easier to prepare my
implementation proposal.

For the second architecture - R-CNN I think I need to read some articles to
have better understanding. I will read [1], [2] and [3] for the weekend and
I will start thinking about the implementation.

I will let know after weekend if I have some questions and concerns.

Thanks,
Ewelina

------------------------------------------------------------
[1] https://arxiv.org/pdf/1311.2524v5.pdf
[2] https://arxiv.org/pdf/1504.08083.pdf
[3] https://arxiv.org/pdf/1506.01497v3.pdf

2018-03-07 21:56 GMT+01:00 Marcus Edel <marcus.edel at fu-berlin.de>:

> Hello Ewelina,
>
> welcome and thanks for getting in touch.
>
> My name is Ewelina Nowak and I am 2nd-year student of Computer Science at
> Gdansk
> University of Technology, Poland. I have experience in ML area, for
> example:
> measuring heart-rate with EEG signals using several ML techniques
> (publication),
> recognition and classification music mood in real-time (thesis from my
> first
> field of study), drone detection using camera and microphone arrays
> (projects
> done at my internships). I am currently at internship at Intel Nervana
> which
> helps develop my skills and experience in the AI area.
>
>
> Sounds like you already looked into some really interesting areas.
>
> Could you please give me more information if any of proposed architectures
> can
> be interesting and useful in mlpack? I would be very grateful for any help
> and
> hints.
>
>
> Each idea is definitely interesting and would fit into the existing
> codebase, my
> recommendation at this point is to focus on one or two ideas. (BRNN isn't
> enough
> for the summer as a single project, but this would be a neat addition).
>
> Let me know if I should clarify anything.
>
> Thanks,
> Marcus
>
> On 7. Mar 2018, at 16:53, Ewelina Nowak <ewelina.anna.nowak15 at gmail.com>
> wrote:
>
> Hello
>
>
> My name is Ewelina Nowak and I am 2nd-year student of Computer Science at
> Gdansk University of Technology, Poland. I have experience in ML area, for
> example: measuring heart-rate with EEG signals using several ML techniques
> (publication), recognition and classification music mood in real-time
> (thesis from my first field of study), drone detection using camera and
> microphone arrays (projects done at my internships). I am currently at
> internship at Intel Nervana which helps develop my skills and experience in
> the AI area.
>
> Recently, I have started to familiarize myself with mlpack code. I
> downloaded mlpack, compiled it from source and set up a development
> environment. Currently I am reading mlpack tutorials and I try to
> experiment with some mlpack ML implementations to get better understanding
> of the project.
>
> I am interested in participating in GSoC 2018 and I am particularly
> interested in Essential Deep Learning Modules. After reading proposed
> papers and after doing some research I have three propositions for ANN
> architectures in which I am interested:
>
> 1. RCNN (Recurrent Convolutional Neural Networks): I have an experience in
> using RNN (with LSTM and GRU units) with CNN in one of my projects: Music
> mood classification using deep learning modules. I used images from
> spectral analysis for predicting mood of a song in time.
>
> 2. R-CNN (Regional based Convolutional Neural Networks): This architecture
> can be used for object detection and classification. It can be used for
> modern modifications of R-CNN: Fast R-CNN, Faster R-CNN or for example Mask
> R-CNN.
>
> 3. BRNN (Bidirectional Recurrent Neural Networks): As I previously
> mentioned I have experience in Recurrent Neural Networks and I would be
> interested in implementing BRNN for one of my new projects in text analysis
> area.
>
> Could you please give me more information if any of proposed architectures
> can be interesting and useful in mlpack? I would be very grateful for any
> help and hints.
>
>
> Best wishes,
>
> Ewelina Nowak
>
> _______________________________________________
> 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/20180309/9001fee8/attachment.html>