 |
ML Newsnew | past | best | rss | submit | about |
| One weird trick for parallelizing convolutional neural networks (arxiv.org) | |
| 3 points by signa11 on Jul 24, 2014 | hide | past | pdf | discuss on HN | |
|
In plain words: A new way to divide convolutional network training across several graphics cards, rather than splitting the work the usual way, so the whole job finishes sooner. It speeds up as more cards are added far better than any other approach. AbstractI present a new way to parallelize the training of convolutional neural networks across multiple GPUs. The method scales significantly better than all alternatives when applied to modern convolutional neural networks. | |
| add comment on HN |