[mlpack] Contribution to mlpack - GSoC '16

[Ryan Curtin]

On Wed, Mar 09, 2016 at 05:11:06AM +0530, Deep Jahan Grewal wrote:
> Hey, 
> 
> I am Deep Jahan Grewal, currently pursuing B.Tech from International
> Institute of Information Technology, Hyderabad, India. I have some
> basic experience in Machine Learning (like successfully understood and
> implemented various algorithms like LDA, PCA, Kernal LDA & PCA,
> Bayesian Classifiers, some clustering techniques, some basic neural
> networks etc). I also have done a group project in which we compared
> different Speech Recognition algorithms including GMM, HMM, RNN and
> Vector Quantization. Also, I have good understanding of Linear
> Programming and convexity and duality associated with it. I am
> currently doing a course on it. I am interested in projects involving
> Semidefinite Programming i.e, 
> 
> 
>     1. LMNN with LRSDP implementation 
>     2. Fixes to MVU low-rank semidefinite programs 
> 
> I have already gone through these materials so that I have some basic
> understanding of semidefinite programming: 
> 
> 
>     1. Introduction to Semidefinite Programming (SDP) by Robert M. Freund 
>     2. Fast Low-Rank Semidefinite Programming for Embedding and Clustering 
> 
> Just to clarify, during the application period, I need to go through
> the literature and then get familiar with MVU implementation or LRSDP
> API of mlpack and if possible, get started with the implementation.
> What should be the main focus right now? 
> 
> It would be very helpful if someone would guide me through the
> specified projects. 

Hi Deep,

I'm very sorry for not responding to this sooner.  Apparently I missed
this email in my inbox somehow.

Both of the projects you've suggested are very difficult.  I would
suggest that you take a look at the following papers:

http://people.orie.cornell.edu/miketodd/orie6327/007-lowrank.pdf
http://www.optimization-online.org/DB_FILE/2003/09/740.pdf
http://www.aaai.org/Papers/AAAI/2006/AAAI06-280.pdf
http://www.cs.cornell.edu/~kilian/papers/171.pdf
http://fodava.gatech.edu/files/reports/FODAVA-09-10.pdf

I know that's a lot to read, but this project requires a good amount of
familiarity with the literature.

When you prepare your proposal, be sure to outline the steps you will
take clearly; although both of these are more research-oriented project,
Google requires some code to be produced over the summer, so we need to
ensure that there are clear deliverables in the project.

Thanks,

Ryan

-- 
Ryan Curtin    | "For more enjoyment and greater efficiency,
ryan at ratml.org | consumption is being standardized."