4 ms·Speeding up neural network training with BLAS and GPUs [pdf]2 points by henning 17y agohenning 17y agoThe authors report speedups of a factor of 3-4 by experimenting with BLAS and GPU-based convolutional neural network implementations in C++.