Y
Hacker News
new
|
ask
|
show
|
jobs
Fast Sparse ConvNets
(
arxiv.org
)
4 points
by
Marat_Dukhan
2402 days ago
1 comments
ekelsen
2402 days ago
TLDR; Make 1x1 convolutions sparse, write fast Sparse Matrix Multiplication kernels, get a nearly 2x speedup with smaller models.
link