[mlpack] Google Summer of Code - 2018

Thu Mar 22 21:08:07 EDT 2018

Hello Hassan,

thanks for the update.

> There is another paper "Text to Photo-realistic Image Synthesis with Stacked
> Generative Adversarial Networks" and its implementation is available. Although,
> the implementations are in python, but this will help me greatly to understand
> the workflow.  I already had knowledge about Bidirectional Recurrent Neural
> Networks. I read the paper and did some information research for proper
> understanding. I will be implementing it too.

Sounds like a good plan to me.

> Is it fine to work on just two modules and produce good results (State of the
> Art accuracy as per the published paper)?

Absolutely, let's focus on a few ideas and make them as fast/usable as possible,
I think backed by reasonable results this would make a great project.

> Regarding the other two projects, can you give me some advice? What else should
> I do next regarding the "Essential Deep Learning Modules"?

For each project there are interesting discussions in the mailing list archive,
so make sure to check the archive. See http://mlpack.org/gsoc.html for
information about how to search the archive. Also, it's possible to submit
multiple ideas but my recommendation is to select a single idea and focus on
that one.

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

Thanks,
Marcus

> On 22. Mar 2018, at 17:36, Hassan Mahmood <14besehmahmood at seecs.edu.pk> wrote:
> 
> Hello Marcus!
> I showed interest in "Essential Deep Learning Modules" project on the mailing list. 
> I digged deep in the project, specifically "Stacked Generative Adversarial Networks" and "Bidirectional Recurrent Neural Networks".
> 
> So far, I have done the following:
> I developed the understanding of mlpack code base by building it on my laptop and used some available functionalities like mlpack_linear_regression and mlpack_perceptron. 
> I went through the neural network code part (Artificial Neural Network (ann)) in mlpack code base and I will be using those existing layers and activation functions.
> I skimmed through Kris Singh's implementation of GAN from GSOC 2017 which will be useful while implementing Stacked GANs. I might use some of his code.
> I also read the Stacked GANs paper and understood the main concept. I just have one or two vague concepts which I am trying to understand.
> There is another paper "Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks" and its implementation is available. Although, the implementations are in python, but this will help me greatly to understand the workflow. 
> I already had knowledge about Bidirectional Recurrent Neural Networks. I read the paper and did some information research for proper understanding. I will be implementing it too.
> 
> Besides, I am interested in "Automatic Bindings to new Languages" and "String Processing Utilities" projects. 
> I am now familiar with code base and workflow for Automatic Bindings procedure in mlpack.
> 
> For now, I am reading and watching tutorials on SGANs and some of its effective variants (changed hyperparameters and optimization). 
> 
> I have few questions:
> Is it fine to work on just two modules and produce good results (State of the Art accuracy as per the published paper)?
> Regarding the other two projects, can you give me some advice?
> What else should I do next regarding the "Essential Deep Learning Modules"?
> 
> I have started working on my application. I will share it with you soon and I would appreciate your feedback.
> 
> Thanks.
> 
> Hassan Mahmood 
> 
> On Sun, Mar 11, 2018 at 10:54 PM, Marcus Edel <marcus.edel at fu-berlin.de <mailto:marcus.edel at fu-berlin.de>> wrote:
> Hello Hassan,
> 
> welcome and thanks for getting in touch.
> 
>> I have implemented YOLO detector, SSD and Segnet for segmentation and have
>> achieved an accuracy of 89% using YOLO and 67% using SSD in real time.
>> Currently, I am working to combine Segnet for segmenting the vehicles and
>> feeding them to Detection Net.
> 
> That sounds really interesting, the capabilities of YOLO are really good, I
> think you can get about 50-60 FPS on a decent GPU?
> 
>> I have a vivid concept about the implementation of the project. I have worked in
>> Python for implementing Convolution Neural Networks, mostly. However, I am
>> proficient in C++ too. So, I am confident that I will be able to complete it in
>> summer.
>> 
>> I would like to have a detail discussion with you on this project.
> 
> Please don't hesitate to ask questions about the project either over the mailing
> list or on IRC (#mlpack).
> 
>> Regarding Variational Autoencoders project, I do not have much of the expertise
>> in generative models but I have used MNIST dataset for learning generative
>> models.  I have skim through the 'Tutorial on Variational AutoEncoders' paper
>> which includes the detailed mathematical explanation of the proposed model. My
>> idea is that this project is research oriented.
> 
> At this point we are mostly interested in the standard implementation, it should
> be possible to use the existing classes (FFN, linear layer); but I agree this
> project could include a research component.
> 
>> Please do let me know, what is expected from me, for the project proposal? What
>> kind of skills, in particular, do I need to exhibit in my proposal? Suggestions
>> regarding how should I proceed with the project or any alternative approach that
>> you have in mind?
> 
> The Application Guide (https://github.com/mlpack/mlpack/wiki/Google-Summer-of- <https://github.com/mlpack/mlpack/wiki/Google-Summer-of->
> Code-Application-Guide) should be useful here.
> 
> I hope anything I said was helpfu, let me know if I should clarify anything.
> 
> Thanks,
> Marcus
> 
>> On 8. Mar 2018, at 18:50, Hassan Mahmood <14besehmahmood at seecs.edu.pk <mailto:14besehmahmood at seecs.edu.pk>> wrote:
>> 
>> Hello there,
>> I am a final year Software Engineering student at National University of Science and Technology, Islamabad.
>> For the past year, I have been working as a Research Assistant at TUKL Lab - SEECS, NUST on various projects related to Deep Learning, Computer Vision and Image Processing. 
>> 
>> Under the supervision of Dr. Faisal Shafait, a reputable professor and a researcher in the field of Computer Vision and Machine learning, I am working on Real Time Vehicle Detection and Recognition as my Senior Year Project. Up till now, I have implemented YOLO detector, SSD and Segnet for segmentation and have achieved an accuracy of 89% using YOLO and 67% using SSD in real time. 
>> Currently, I am working to combine Segnet for segmenting the vehicles and feeding them to Detection Net.
>> 
>> I am writing this email to show an interest in the project "Essential Deep Learning Modules" for Google Summer of Code 2018. I have good expertise and domain knowledge of Deep Learning Models. It aligns with my research area and my senior year project. I feel excited about the project and have a real hope to work on it.
>> It will greatly help me in my PhD admission and my career, too. 
>> 
>> I have a vivid concept about the implementation of the project. I have worked in Python for implementing Convolution Neural Networks, mostly. However, I am proficient in C++ too. So, I am confident that I will be able to complete it in summer.
>> 
>> I would like to have a detail discussion with you on this project. 
>> 
>> I also find "Variational Autoencoders" and "String Processing Utilities" projects, interesting. 
>> 
>> Regarding Variational Autoencoders project, I do not have much of the expertise in generative models but I have used MNIST dataset for learning generative models. 
>> I have skim through the 'Tutorial on Variational AutoEncoders' paper which includes the detailed mathematical explanation of the proposed model. My idea is that this project is research oriented.
>> 
>> Please do let me know, what is expected from me, for the project proposal? What kind of skills, in particular, do I need to exhibit in my proposal? Suggestions regarding how should I proceed with the project or any alternative approach that you have in mind?
>> 
>> I would really appreciate, if you can help me out.
>> 
>> Thank you.
>> 
>> Regards,
>> Hassan Mahmood
>> Software Engineering
>> NUST H-12 Islamabad
>> 
>> _______________________________________________
>> mlpack mailing list
>> mlpack at lists.mlpack.org <mailto:mlpack at lists.mlpack.org>
>> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack <http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack>
> 
> 
> 
> -- 
> Hassan Mahmood
> Software Engineering
> NUST H-12 Islamabad
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20180323/f67f5bf1/attachment-0001.html>