Скачать или смотреть Fast Algorithms for Convolutional Neural Networks by Andrew Lavin and Scott Gray

This is a reading group talk on the published paper in CVPR 2016 entitled, "Fast Algorithms for Convolutional Neural Networks," by Andrew Lavin and Scott Gray.

http://arxiv.org/abs/1509.09308

It's a super rough introduction on what *I think* this paper is about. Note: I am not a co-author of this paper, this video is here in the hopes that someone else will benefit from the slides I generated for our reading group in the Vision group at the University of Washington.

I ran out of battery at the end. But to end, we can take advantage of Hermitian symmetry for real FFTs and only use a subset of the complex valued terms.

Here's some useful references that I used to make this presentation:

1.   / fast_algorithms_for_convolutional_neural_n...  
2.   / nnpack_acceleration_package_for_neural_net...  
3. https://www.nervanasys.com/winograd/
4. http://arxiv.org/abs/1412.7580
5. https://arxiv.org/abs/1312.5851

Slides here: http://homes.cs.washington.edu/~cdel/...

Carlo
http://homes.cs.washington.edu/~cdel/