NEW: RESOURCE PAGE

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!

Wednesday, March 11, 2015

OctaPi seen chasing Cuda cat!

Before moving from Google to Baidu, Andrew Ng built an image processing cat detector, training a net to run on a large network of standard PCs. And then he got the idea that it could be done on a budget for about $20000 of hardware among which a bunch of GPUs: Neural net computations are a good fit for the massively parallel SIMD (single instruction multiple data)  processor which helps game players get fast frame rates. 

Unsurprisingly, Nvidia quite likes this sort of scientific application, as it helps sell more graphics chips, and  maybe even helps makes their product an specialised component indispensable of tomorrow's smart devices.

Enter CUDA, the development library Nvidia supplies to maintain a measure of device independence, and help developers get the most out of GPUs;  a sponsored course on Udacity will help programmers attempt this 50x speedup and those impossible Teraflop speeds ...

But wait! Anyone has a GPU these days! With a bit of Google help, I even managed to locate the HOWTO for doing CUDA-enabled work on my Macbook Pro.  There's a Cuda drop in for Octave, for R, for Matlab I believe, and for encog  etc.

I'm a bit sceptical about notebook CUDA performance, but if the setup is easy enough I'm game for a test. And I guess, if you're desperate to spin up a cluster faster than you can say "Lawrence Livermore",  there's always Amazon EC2

There is a very interesting Nvidia page on Deep Learning in neural nets, which links to some hard to locate machine vision papers.  And I guess if you don't 'ave the budget for Nvidia's products you can always rely on the Raspberry Pi's embedded graphics processor - GCHQ will show you how it's done: 

GCHQ builds monster Raspberry Pi cloud with OctaPi formation

UPDATE a random link: https://timdettmers.wordpress.com/2014/08/14/which-gpu-for-deep-learning/

PS. My Macbook Pro's CPU/FPU does 120GFlops! Miaoooouuuu!

Edmund

No comments:

Post a Comment

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