NAS-Bench-1Shot1
收藏arXiv2020-04-13 更新2024-06-21 收录
下载链接：
https://github.com/automl/nasbench-1shot1
下载链接
链接失效反馈官方服务：
资源简介：
NAS-Bench-1Shot1是由弗莱堡大学计算机科学系创建的一个用于神经架构搜索（NAS）的基准数据集。该数据集包含363,648个独特的架构，旨在通过提供一个大规模的表格基准，支持对一击式NAS方法的廉价评估。数据集的创建过程涉及对搜索空间的创新映射，允许在不重新训练的情况下查询找到的架构性能，从而支持对搜索过程的细致和统计上可靠的分析。NAS-Bench-1Shot1的应用领域主要集中在提高NAS方法的效率和可重复性，解决现有NAS研究中存在的可重复性和成本问题。
NAS-Bench-1Shot1 is a benchmark dataset for neural architecture search (NAS) developed by the Department of Computer Science, University of Freiburg. It contains 363,648 unique architectures, and is designed to support low-cost evaluation of one-shot NAS methods by providing a large-scale tabular benchmark. The dataset creation involves an innovative mapping of the NAS search space, which allows querying the performance of discovered architectures without retraining, thereby enabling detailed and statistically rigorous analysis of the search process. The primary application scenarios of NAS-Bench-1Shot1 focus on enhancing the efficiency and reproducibility of NAS methods, addressing the reproducibility and cost-related issues prevalent in current NAS research.
提供机构：
弗莱堡大学计算机科学系
创建时间：
2020-01-28
