BriefGPT.xyz
Dec, 2022
FSCNN:一种快速的稀疏卷积神经网络推理系统
FSCNN: A Fast Sparse Convolution Neural Network Inference System
HTML
PDF
Bo Ji, Tianyi Chen
TL;DR
本文主要介绍了利用大规模稀疏结构提高卷积神经网络推理效率的方法, 开发了一种名为 FSCNN 的稀疏卷积神经网络推理系统,能够在一定的稀疏程度下优于标准深度学习库 PyTorch,但遇到高优化密集运算时推理效率有一定局限性,因此推荐采用粗粒度稀疏度进行通用模型压缩。
Abstract
convolution neural networks
(CNNs) have achieved remarkable success, but typically accompany high computation cost and numerous redundant weight parameters. To reduce the FLOPs,
structure pruning
is a popular app
→