arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2203.06905cs.LGcs.AIcs.CV
收藏

少即是多:NAS方法中的代理数据集

Less is More: Proxy Datasets in NAS approaches

German Research Center for Artificial Intelligence (DFKI)(德国人工智能研究中心(DFKI));TU Kaiserslautern(凯泽斯劳滕工业大学)

Brian Moser, Federico Raue, Jörn Hees, Andreas Dengel

AI总结:

本文在NAS-Bench-201上通过约1400次实验,系统分析了代理数据集对NAS搜索的影响,发现将训练数据减少至25%可保持同等精度并缩短搜索时间,且部分子集设计准确率可提升22个百分点。

摘要:

神经架构搜索(NAS)将神经网络的设计定义为一个搜索问题。不幸的是,由于设计中的元素数量和它们之间可能的连接方式存在多种可能性,NAS的计算量非常大。在这项工作中,我们基于几种减少数据集大小的采样方法(无监督和有监督情况)广泛分析了数据集大小在减少搜索时间方面的作用,作为一种与具体NAS方法无关的通用方法。我们将这些技术与NAS-Bench-201中的四种常见NAS方法进行了比较,在CIFAR-100上进行了约1,400次实验。我们一个令人惊讶的发现是,在大多数情况下,我们可以将训练数据量减少到25%,从而将搜索时间也减少到25%,同时保持与在完整数据集上训练相同的准确率。此外,从子集导出的某些设计在准确率上比从完整数据集导出的设计高出多达22个百分点。

英文摘要:

Neural Architecture Search (NAS) defines the design of Neural Networks as a search problem. Unfortunately, NAS is computationally intensive because of various possibilities depending on the number of elements in the design and the possible connections between them. In this work, we extensively analyze the role of the dataset size based on several sampling approaches for reducing the dataset size (unsupervised and supervised cases) as an agnostic approach to reduce search time. We compared these techniques with four common NAS approaches in NAS-Bench-201 in roughly 1,400 experiments on CIFAR-100. One of our surprising findings is that in most cases we can reduce the amount of training data to 25\%, consequently reducing search time to 25\%, while at the same time maintaining the same accuracy as if training on the full dataset. Additionally, some designs derived from subsets out-perform designs derived from the full dataset by up to 22 p.p. accuracy.

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。

↑