Re:Facebook open-sources deep-learning modules
by Hydraulix989 on Friday January 16, 2015 @03:15PM
(#8901391)
In my experience, the fully connected layers are the bottleneck. The other issue was the alternating compute-heavy convolution and the IO-heavy pooling. I'm curious how this FFT implementation stacks up against cuDNN (what's the speedup like for just the convolutional layers? and then what's the overall speedup like?).
[ Parent
| on HN ]
Re:Facebook open-sources deep-learning modules
by ajtulloch on Friday January 16, 2015 @03:39PM
(#8901531)
http://arxiv.org/pdf/1412.7580v2.pdf compares the convolutional implementation with the cuDNN layers. For the FC layers, it's just CuBLAS `sgemm`.