发表机构
School of Computer Science and Engineering, Nanyang Technological University; HP-NTU Digital Manufacturing Corporate Lab, Nanyang Technological University; Department of Computer Science, Norwegian University of Science and Technology（南洋理工大学计算机科学与工程学院; 惠普-南洋理工大学数字制造联合实验室; 挪威科技大学计算机科学系）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究针对嵌入式硬件的高效卷积神经网络，提出TECO多维剪枝框架，通过两阶段重要性评估框架全面评估剪枝单元，用启发式算法逐步修剪网络的深度、宽度和分辨率，实验验证其优于现有方法。
AI 中文摘要
本文提出了TECO，一个多维剪枝框架，用于协同修剪卷积神经网络（CNN）的三个维度（深度、宽度和分辨率），以在嵌入式硬件上实现更好的执行效率。在TECO中，首先引入两阶段重要性评估框架，根据各维度内的局部重要性和跨维度的全局重要性，有效且全面地评估每个剪枝单元。基于该评估框架，提出启发式剪枝算法，逐步修剪CNN的三个维度，以在准确性和效率之间实现最佳权衡。在多个基准上进行的实验验证了TECO优于现有最先进（SOTA）方法。代码和预训练模型可在指定网址获取。
英文摘要
In this paper, we propose TECO, a multi-dimensional pruning framework to collaboratively prune the three dimensions (depth, width, and resolution) of convolutional neural networks (CNNs) for better execution efficiency on embedded hardware. In TECO, we first introduce a two-stage importance evaluation framework, which efficiently and comprehensively evaluates each pruning unit according to both the local importance inside each dimension and the global importance across different dimensions. Based on the evaluation framework, we present a heuristic pruning algorithm to progressively prune the three dimensions of CNNs towards the optimal trade-off between accuracy and efficiency. Experiments on multiple benchmarks validate the advantages of TECO over existing state-of-the-art (SOTA) approaches. The code and pre-trained models are available at https://github.com/ntuliuteam/Teco.
CommentsAuthor's accepted version. Published in Proceedings of the 60th ACM/IEEE Design Automation Conference (DAC 2023)
Journal refProceedings of the 60th ACM/IEEE Design Automation Conference (DAC), pp. 1-6, 2023
DOI：10.1109/DAC56929.2023.10247965
