一种剪枝搜索的模型压缩方法

By evaluating and grouping pruning ratios using an adaptive search method and optimizing shared parameters across all channels, the problem of deep learning model compression under diverse hardware conditions on embedded platforms is solved, achieving efficient model adaptability and resource utilization.

CN115829021BActive Publication Date: 2026-07-17SOUTH CHINA UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SOUTH CHINA UNIV OF TECH
Filing Date
2022-11-08
Publication Date
2026-07-17

Smart Images

  • Figure CN115829021B_ABST
    Figure CN115829021B_ABST
Patent Text Reader

Abstract

本发明公开了一种剪枝搜索的模型压缩方法，包括以下步骤：S1、输入图像数据，建立卷积神经网络模型，对卷积神经网络模型进行搜索训练，优化卷积神经网络模型的权重参数；S2、对卷积神经网络模型每个块评估块对误差的敏感性，根据敏感性将块进行分组，产生剪枝比例配置候选集；S3、对卷积神经网络模型进行最大剪枝比例、最小剪枝比例和在配置集中随机采样剪枝比例下的优化，训练得出全通道共享参数的网络；S4、根据应用场景的硬件约束在全通道共享参数的网络中选择合适的剪枝比例配置。S5、对剪枝网络中的批标准化层的统计参数进行校准。
Need to check novelty before this filing date? Find Prior Art