Faster deep learning with GPUs and Theano

blog.dominodatalab.com

41 pointsgk14 comments

Using GPUs for deep learning is like, step 1. This is really well known, and has been since the first big papers on the topic.

Hinton's "ImageNet Classification with Deep Convolutional Neural Networks" has the word "GPU" in it 34 times.

http://www.cs.toronto.edu/~fritz/absps/imagenet.pdf

The point of this article is to describe exactly how to do this using a specific Python library. It's not fundamental research but it's very useful for people who want to do this sort of thing in practice without reinventing all the software libraries from scratch.

This one is very helpful, too: http://danielnouri.org/notes/2014/12/17/using-convolutional-...

If you're going for speed, here's some benchmarks of different libraries on ImageNet and convolutional layers: https://github.com/soumith/convnet-benchmarks

Reply on news.ycombinator.com