Single Path One-Shot Neural Architecture Search with Uniform Sampling

dyweb / papers-notebook

:page_facing_up: :cn: :page_with_curl: 论文阅读笔记（分布式系统、虚拟化、机器学习）Papers Notebook (Distributed System, Virtualization, Machine Learning)

https://github.com/dyweb/papers-notebook/issues?utf8=%E2%9C%93&q=is%3Aissue+is%3Aopen+-label%3ATODO-%E6%9C%AA%E8%AF%BB

Apache License 2.0

2.12k stars 244 forks source link

Single Path One-Shot Neural Architecture Search with Uniform Sampling #140

Open gaocegege opened 5 years ago

gaocegege commented 5 years ago

https://arxiv.org/abs/1904.00420

gaocegege commented 5 years ago

https://zhuanlan.zhihu.com/p/63605721

gaocegege commented 5 years ago

跟 #143 异曲同工，但是它是用 super kernel 来做的路径压缩

用一个7x7的大卷积，来代表3x3,5x5,7x7的三种卷积，把外边一圈mask清零掉就变成了3x3或5x5，这个大的卷积成为 superkernel, 于是整个网络只有一种卷积了，看起来是一个直筒结构。这里还有一点weight sharing 的味道，最里面的，比如3x3的weights是和5x5,7x7的卷积共享的。