[mlpack] GSoC 2016 - Project Questions

[Yannis Mentekidis]

Hello developers,

My name is Yannis Mentekidis, I am a final-year student in Electrical and
Computer Engineering in Aristotle University, Thessaloniki Greece.

I've gone through your project description and your ideas page, and I would
love to work with you for the summer! I've started going through your
source code and compiled the library to familiarize myself with the project.

Two projects are especially relevant to my interests: Approximate Nearest
Neighbors and Neuroevolutional Algorithms.

I've worked for my thesis on efficient algorithms for the Nearest Neighbor
Search problem, and I am familiar with Approximate NNS – for example, LSH
and Multiprobe LSH, as well as Compressed Hashing (from CVPR 2013). I would
be interested in implementing some or all of these, but I would also like
to try the ideas presented in “Modeling LSH for performance tuning (Dong et
al, 2008)”. You also mention dual trees, which I am less familiar with.

*My first question is*, would some such implementation (Multiprobe
LSH/Performance Tuning for LSH) be in the interest of the project? Or is it
preferable to design a larger number of approximate solutions and give more
freedom of choice to the end user?

*Another question* is inspired from an old ticket I found in GitHub, where
you mention interest in parallelization. I would also be interested in, as
a side goal, implementing a parallel version of a few of the approximate
algorithms. Is that something interesting to you guys?

The other subject, Neuroevolutional Algorithms, I have less experience on,
but I would like to work on evolutionary computation and I have in the past
implemented feedforward Neural Networks, so this is also something I could
see myself working on.

*My final question* is regarding procedure: Since I am interested in both
these projects, should I submit two different applications to Google, or
just one where I mention both projects?

I will be lurking your irc channel and GitHub in the following weeks,
hopefully we will have the chance to work together!

Regards,

Yannis
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cc.gatech.edu/pipermail/mlpack/attachments/20160302/3c564a24/attachment-0002.html>