一种基于通道相似度的剪枝方法、终端及可读存储介质
By calculating the similarity between channels and channel sets and resolving the dependency graph, the importance of channels is evaluated for precise pruning, which solves the problem of insufficient pruning in existing technologies and improves the performance and efficiency of neural network models.
CN117910534BActive Publication Date: 2026-07-17SHENZHEN UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN UNIV
- Filing Date
- 2023-12-27
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing techniques fail to adequately consider the intrinsic relationships between channels when pruning neural network models, resulting in a significant decrease in model performance after pruning.
Method used
By calculating the similarity between channels and channel sets, the importance of channels is evaluated using representation vectors, and the network structure is analyzed using dependency graphs to perform precise channel pruning.
Benefits of technology
This improved pruning precision, avoided pruning critical channels, and ensured the stability and efficiency of the model's performance after pruning.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN117910534B_ABST
Abstract
本发明公开一种基于表征向量的通道剪枝方法、终端及可读存储介质，所述方法包括：获取神经网络模型各个通道对应的集合的特征，通道对应的集合为包含该通道以外的所有通道的集合；分别计算各个所述通道的特征和与通道对应的集合的特征之间的相似度；根据所述相似度对所述神经网络模型进行剪枝，得到剪枝后的神经网络模型。本发明通过计算通道与通道的集合之间的相似度，作为该通道重要性的评估，这种方式相较现有技术通过对通道计算两两之间的相似度并进行求和的方式，更充分地考虑了通道之间的联系，具有更高的剪枝精度。
Need to check novelty before this filing date? Find Prior Art