Convolutional Neural Network on a GPU

Dropout CNN trained with Adadelta on the MNIST dataset

Posted by Matt on October 11, 2014

The documentation that exists today with respect to getting a GPU up and running and detailed analysis of convet architecture is so thorough and well explained that anyone following my old post would probably be doing themselves a diservice as well as installing a bunch of old gpu software. I decided to just scrap the post altogether and save the re-write for another time.

Since you're already here..

If you came to this blog to legimately learn how CNNs work, just head here: http://colah.github.io/posts/2014-07-Understanding-Convolutions/ or here: http://deeplearning.net/tutorial/lenet.html

If that didn't satisfy you, try some of these:

www.arxiv-sanity.com

http://gitxiv.com

http://karpathy.github.io

https://github.com/fchollet/keras/tree/master/examples

https://github.com/torch/demos

https://github.com/dmlc/mxnet/tree/master/example

https://www.tensorflow.org/versions/r0.7/tutorials/index.html

If all of this is entirely confusing to you and you're looking for somewhere in this weird science to lay your roots, check out this old classic: Bayesian Methods for Hackers. It's better to start here with a strong foundation for the 99.9% of the problems you'll ever face than to try to deep learn all the things.

Check out some of the stuff we do