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 2101.09808cs.LG
收藏

CNN优化的分析表征和设计空间探索

Analytical Characterization and Design Space Exploration for Optimization of CNNs

University of Utah(犹他大学);Washington State University(华盛顿州立大学);Ohio State University(俄亥俄州立大学)

Rui Li, Yufan Xu, Aravind Sukumaran-Rajam, Atanas Rountev, P. Sadayappan

摘要:

在存储层次结构中移动数据是一个基本瓶颈,可能会限制机器学习核心算法的性能,例如卷积神经网络(CNN)。循环级优化,包括循环平铺和循环置换,是减少数据移动的基本转换。然而,用于找到最佳循环级优化配置的搜索空间是爆炸性的大。本文开发了一种分析建模方法,用于在多核CPU上找到CNN的最佳循环级优化配置。实验评估表明,这种方法实现了与最先进的库和基于自动调整的CNN优化器相当或更好的性能。

英文摘要:

Moving data through the memory hierarchy is a fundamental bottleneck that can limit the performance of core algorithms of machine learning, such as convolutional neural networks (CNNs). Loop-level optimization, including loop tiling and loop permutation, are fundamental transformations to reduce data movement. However, the search space for finding the best loop-level optimization configuration is explosively large. This paper develops an analytical modeling approach for finding the best loop-level optimization configuration for CNNs on multi-core CPUs. Experimental evaluation shows that this approach achieves comparable or better performance than state-of-the-art libraries and auto-tuning based optimizers for CNNs.

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。

↑