Searching for tutorials and software about Deep Learning and Neural Nets? Be sure to look at my Resource Page!
Looking for Octave? Go to my Easy Octave on Mac page!

Friday, April 17, 2015

From reproduction of research to original research

I am at a decision point concerning my next time investment.

By modding Andrea's code and getting decent scores for in a few minutes for the MNIST examples, I believe I've confirmed that I can use my laptop to do neural net work at the state of the art 15 years ago. At this point I'm basically back to where I was when I finished my PhD.

I could now move from the antiquated and simple FANN shallow net framework to a more sophisticated environment like Theano, Pylearn2, Torch or Caffe, and tackle harder problems with more recent methods. 

However I'm afraid of two things:
  • A slow learning curve as I wade my way through a novel programming environment.
  • Inadequate computational resources: my laptop vs. multi-GPU clusters at Google and M$.  I am competing with clusters that are 3 orders of magnitude faster. 
The alternatives are thus to either learn modern methods and locate some powerful equipment —eg. an Amazon AWS instance with a GPU—  or to think up some original contributions which can be usefully demonstrated with an old set of tools and benchmarks.

Comments are welcome. 


No comments:

Post a Comment

Hey, let me know what you think of my blog, and what material I should add!