Editor's Note: Here is a list of references compiled by Tim Spann for using CaffeOnSpark for distributed deep learning. Three references to GitHub, one to Yahoo!, and another to Berkeley Vision.
Yahoo! has done some good work getting the deep learning framework, Caffe, to work on Spark (including Spark on YARN).
Getting Caffe working takes a bit of work to get Python and all it's dependencies installed. You need to follow the install guide carefully and stick with a supported OS.