BriefGPT.xyz
Sep, 2019
平衡一次性神经架构优化
Understanding and Improving One-shot Neural Architecture Optimization
HTML
PDF
Renqian Luo, Tao Qin, Enhong Chen
TL;DR
通过平衡神经网络结构大小的训练,提出了Balanced NAO方法来解决One-shot NAS性能劣于传统NAS的问题,该方法的综合实验表明其有效性和鲁棒性,通过该方法在CIFAR-10和ImageNet数据集上发现的架构较基线模型可以将测试误差率降低到2.60%和74.4%的top-1准确率。
Abstract
The ability of accurately ranking candidate architectures is the key to the performance of
neural architecture search
~(NAS).
one-shot nas
is proposed to cut the expense but shows inferior performance against conv
→