BriefGPT.xyz
Jul, 2020
通过结构化连续稀疏化增加深度网络的效率
Growing Efficient Deep Networks by Structured Continuous Sparsification
HTML
PDF
Xin Yuan, Pedro Savarese, Michael Maire
TL;DR
本文提出了一种基于准确性和稀疏性目标的深层网络动态构建算法,与传统的剪枝方法不同,本方法采用渐进式连续松弛和网络优化,然后采样稀疏子网络,使得训练出来的深层网络更加高效。实验结果证明,采用本算法训练的网络比其他竞争的剪枝方法更加精确且规模更小。
Abstract
We develop an approach to
training
deep networks
while dynamically adjusting their
architecture
, driven by a principled combination of acc
→