arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2212.03354cs.LGcs.ARcs.NEcs.PF
收藏

HADAS:硬件感知的动态神经架构搜索,用于边缘性能扩展

HADAS: Hardware-Aware Dynamic Neural Architecture Search for Edge Performance Scaling

LAMIH/UMR CNRS, Université Polytechnique Hauts-de-France(法国上法兰西综合理工大学 LAMIH/法国国家科学研究中心联合研究单位);University of California, Irvine(加州大学欧文分校)

Halima Bouzidi, Mohanad Odema, Hamza Ouarnoughi, Mohammad Abdullah Al Faruque, Smail Niar

摘要:

动态神经网络(DyNN)已经成为在资源受限的边缘设备上实现智能同时保持计算效率的可行技术。在许多情况下,DyNN的实现可能是次优的,因为其底层骨干架构在设计阶段独立于以下两者而被开发:(i)动态计算特征,例如提前退出,以及(ii)底层硬件的资源效率特征,例如,动态电压和频率缩放(DVFS)。为了解决这个问题,我们提出了HADAS,这是一种新型的硬件感知动态神经架构搜索框架,它实现了DyNN架构,其骨干,早期退出功能和DVFS设置已经过联合优化,以最大限度地提高性能和资源效率。我们使用CIFAR-100数据集和一组不同的边缘计算平台进行的实验表明,与传统的动态模型相比,HADAS动态模型实现了高达57%的能效提升,同时保持了所需的准确度得分水平。我们的代码可在https://github.com/HalimaBouzidi/HADAS上获得

英文摘要:

Dynamic neural networks (DyNNs) have become viable techniques to enable intelligence on resource-constrained edge devices while maintaining computational efficiency. In many cases, the implementation of DyNNs can be sub-optimal due to its underlying backbone architecture being developed at the design stage independent of both: (i) the dynamic computing features, e.g. early exiting, and (ii) the resource efficiency features of the underlying hardware, e.g., dynamic voltage and frequency scaling (DVFS). Addressing this, we present HADAS, a novel Hardware-Aware Dynamic Neural Architecture Search framework that realizes DyNN architectures whose backbone, early exiting features, and DVFS settings have been jointly optimized to maximize performance and resource efficiency. Our experiments using the CIFAR-100 dataset and a diverse set of edge computing platforms have seen HADAS dynamic models achieve up to 57% energy efficiency gains compared to the conventional dynamic ones while maintaining the desired level of accuracy scores. Our code is available at https://github.com/HalimaBouzidi/HADAS

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。

↑