基于批量归一化层的重要性解耦剪枝策略分析
Analysis of Importance Decoupling Pruning Strategy Based on Batch Normalization Layer
裘孜杰 1匡迎春1
作者信息
- 1. 湖南农业大学信息与智能科学技术学院,湖南 410128
- 折叠
摘要
阐述一种通过重要性分析进行解耦的深度学习自动稀疏-剪枝策略,通过重要性判断对神经网络强制进行通道级重要性解耦稀疏训练.该方法准确地解耦冗余通道,最小化剪枝后的精度损失.
Abstract
This paper describes a deep learning automatic sparse pruning strategy that decouples through importance analysis,and forces channel level importance decoupling sparse training on neural networks through importance judgment.This method accurately decouples redundant channels and minimizes accuracy loss after pruning.
关键词
智能技术/模型剪枝/模型压缩/结构化剪枝Key words
intelligent technology/model pruning/model compression/structured pruning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据