一种面向边缘环境的深度学习模型压缩方法
By combining cross-layer feature fusion and distillation loss function, the challenge of deploying deep learning models in edge environments is solved, improving model performance and efficiency while reducing training costs and protecting data privacy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUILIN UNIV OF ELECTRONIC TECH
- Filing Date
- 2024-01-18
- Publication Date
- 2026-07-17
AI Technical Summary
Existing large-scale deep learning models are difficult to deploy in resource-constrained edge environments, and existing knowledge distillation methods neglect training costs and efficiency issues when improving the accuracy of student networks.
A cross-layer feature fusion method is adopted, which aligns deep and shallow features through nearest neighbor upsampling and 1×1 convolution kernels, and combines the cross-layer distillation loss function to directly use the original feature map of the teacher model to perform knowledge distillation and train the student model.
It improves the performance and efficiency of deep learning models in edge environments, reduces training costs, and protects data privacy, making it suitable for resource-constrained edge devices.
Smart Images
Figure CN118135240B_ABST