Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2406.03879 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916465459855360 |
|---|---|
| author | Yang, Minghao Gao, Linlin Li, Pengyuan Li, Wenbo Dong, Yihong Cui, Zhiying |
| author_facet | Yang, Minghao Gao, Linlin Li, Pengyuan Li, Wenbo Dong, Yihong Cui, Zhiying |
| contents | Current structured pruning methods often result in considerable accuracy drops due to abrupt network changes and loss of information from pruned structures. To address these issues, we introduce the Decay Pruning Method (DPM), a novel smooth pruning approach with a self-rectifying mechanism. DPM consists of two key components: (i) Smooth Pruning: It converts conventional single-step pruning into multi-step smooth pruning, gradually reducing redundant structures to zero over N steps with ongoing optimization. (ii) Self-Rectifying: This procedure further enhances the aforementioned process by rectifying sub-optimal pruning based on gradient information. Our approach demonstrates strong generalizability and can be easily integrated with various existing pruning methods. We validate the effectiveness of DPM by integrating it with three popular pruning methods: OTOv2, Depgraph, and Gate Decorator. Experimental results show consistent improvements in performance compared to the original pruning methods, along with further reductions of FLOPs in most scenarios. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_03879 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Decay Pruning Method: Smooth Pruning With a Self-Rectifying Procedure Yang, Minghao Gao, Linlin Li, Pengyuan Li, Wenbo Dong, Yihong Cui, Zhiying Machine Learning Computer Vision and Pattern Recognition Current structured pruning methods often result in considerable accuracy drops due to abrupt network changes and loss of information from pruned structures. To address these issues, we introduce the Decay Pruning Method (DPM), a novel smooth pruning approach with a self-rectifying mechanism. DPM consists of two key components: (i) Smooth Pruning: It converts conventional single-step pruning into multi-step smooth pruning, gradually reducing redundant structures to zero over N steps with ongoing optimization. (ii) Self-Rectifying: This procedure further enhances the aforementioned process by rectifying sub-optimal pruning based on gradient information. Our approach demonstrates strong generalizability and can be easily integrated with various existing pruning methods. We validate the effectiveness of DPM by integrating it with three popular pruning methods: OTOv2, Depgraph, and Gate Decorator. Experimental results show consistent improvements in performance compared to the original pruning methods, along with further reductions of FLOPs in most scenarios. |
| title | Decay Pruning Method: Smooth Pruning With a Self-Rectifying Procedure |
| topic | Machine Learning Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2406.03879 |