AnalogNAS-Bench
收藏资源简介:
AnalogNAS-Bench是一个针对模拟内存计算(AIMC)的神经网络架构搜索(NAS)基准。该数据集旨在解决传统数字硬件在AIMC平台上的性能问题,通过硬件感知训练(HWT)评估架构的鲁棒性。数据集基于NAS-Bench-201的搜索空间,并引入了AIMC特有的硬件非理想性约束。AnalogNAS-Bench为研究人员提供了一个标准化的框架,用于比较和分析不同架构在AIMC条件下的性能。
AnalogNAS-Bench is a neural architecture search (NAS) benchmark for analog in-memory computing (AIMC). This benchmark aims to address the performance issues of conventional digital hardware on AIMC platforms, and evaluates the robustness of neural architectures via Hardware-Aware Training (HWT). It is built upon the search space of NAS-Bench-201, and introduces AIMC-specific hardware non-ideality constraints. AnalogNAS-Bench provides researchers with a standardized framework to compare and analyze the performance of different neural network architectures under AIMC conditions.




