[mlpack] Gsoc 2014

Ryan Curtin gth671b at mail.gatech.edu
Tue Feb 25 12:24:53 EST 2014


On Tue, Feb 25, 2014 at 10:39:08PM +0530, Aakash Chaudhary wrote:
> I want to apply to your org in gsoc 2014
> i have seen the ideas page and i think i could give my best in implement
> tree types
> 
> What i am doing now is get to know mlpack and use it and also i am
> Studying the trees ( ball-trees, kd-trees and all)
> 
> apart from that what should i do?

That sounds good.  It would probably be useful to read many papers on
different types of trees, including the paper that details the
abstractions on which mlpack trees are based:

http://machinelearning.wustl.edu/mlpapers/paper_files/icml2013_curtin13.pdf

And here are some papers on trees.  A good theoretical understanding of
what these trees do and why they work is a necessity to be able to
implement these well.

Metric trees (sorry, no PDF link)
http://www.sciencedirect.com/science/article/pii/002001909190074R

kd-trees (notation may be a little weird, it's a very old paper)
http://cgi.di.uoa.gr/~ad/MDE515/p509-bentley.pdf

cover trees
http://machinelearning.wustl.edu/mlpapers/paper_files/icml2006_BeygelzimerKL06.pdf

k-means tree (sorry, no PDF link)
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1674939

random projection trees
http://homes.mpimf-heidelberg.mpg.de/~mhelmsta/pdf/2007%20Freund%20rpTrees.pdf

Bregman ball trees (a bit less related)
http://cseweb.ucsd.edu/~lcayton/bbtree.pdf

vantage point trees
http://aidblab.cse.iitm.ac.in/cs625/vptree.pdf

> and also how to convince that i am capable of doing all this to you?
> 
> what i can think(convincing you) of is to discuss with you how will i
> implement all this

Figuring out how to write a good application isn't something that I can
tell you.  You could consult the many resources for potential GSoC
students available on the internet, maybe starting here:

http://en.flossmanuals.net/GSoCStudentGuide/

> and last one i didn't see any mentors specified on ideas page so who would
> be mentoring (for implementing tree types)?
> [so that i can contact him for all small details]

That would be me.  I'd prefer to keep communication on the list for the
sake of anyone else who might be perusing the archives, if you don't
mind.

-- 
Ryan Curtin    | "But feel, to the very end, the triumph of being
ryan at ratml.org | alive!"  - Jöns


More information about the mlpack mailing list