Fetching the paper…

FireCaffe: near-linear acceleration of deep neural network training on compute clusters · Around