arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2103.07289cs.CV
收藏

Searching by Generating:灵活高效的一次NAS与架构生成器

Searching by Generating: Flexible and Efficient One-Shot NAS with Architecture Generator

National Cheng Kung University(成功大学)

Sian-Yao Huang, Wei-Ta Chu

摘要:

在one-shot NAS中,需要从超网中搜索子网以满足不同的硬件约束。然而,搜索成本高,并且对于$N$不同的约束需要$N$次搜索。在这项工作中,我们提出了一种新的搜索策略称为架构生成器搜索子网络,通过生成它们,使搜索过程可以更有效和灵活。使用训练的架构生成器,给定目标硬件约束作为输入,$N$良好的架构可以通过仅一次前向传递而无需重新搜索和超网再训练来生成用于$N$约束。此外,我们提出了一种新的单路径超网,称为统一超网,以进一步提高搜索效率,减少GPU的内存消耗的架构生成器。结合架构生成器和统一超网,提出了一个灵活高效的一次性NAS框架,称为生成式搜索NAS(SGNAS)。使用预训练的supernt,SGNAS对$N$不同硬件约束的搜索时间仅为5 GPU小时,比以前的SOTA单路径方法快$4N$倍。在从头开始训练之后,SGNAS在ImageNet上的top1准确率为77.1%,与SOTA相当。该代码可从以下网址获取:https://github.com/eric8607242/SGNAS。

英文摘要:

In one-shot NAS, sub-networks need to be searched from the supernet to meet different hardware constraints. However, the search cost is high and $N$ times of searches are needed for $N$ different constraints. In this work, we propose a novel search strategy called architecture generator to search sub-networks by generating them, so that the search process can be much more efficient and flexible. With the trained architecture generator, given target hardware constraints as the input, $N$ good architectures can be generated for $N$ constraints by just one forward pass without re-searching and supernet retraining. Moreover, we propose a novel single-path supernet, called unified supernet, to further improve search efficiency and reduce GPU memory consumption of the architecture generator. With the architecture generator and the unified supernet, we propose a flexible and efficient one-shot NAS framework, called Searching by Generating NAS (SGNAS). With the pre-trained supernt, the search time of SGNAS for $N$ different hardware constraints is only 5 GPU hours, which is $4N$ times faster than previous SOTA single-path methods. After training from scratch, the top1-accuracy of SGNAS on ImageNet is 77.1%, which is comparable with the SOTAs. The code is available at: https://github.com/eric8607242/SGNAS.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑