New sparse neural network exploration technique “DWA” presented at CIKM 2025 / Research team led by Professor Kim Jang Ho (Department of Artificial Intelligence)
- 25.11.25 / 전윤실
Research team led by Professor Kim Jang Ho (Department of Artificial Intelligence, Kookmin University) (Kim Jin Woo, Shin Jong Yun, Ahn Sang Ho, Kim Jang Ho) presented DWA (Dynamic Pruning with Weight Alignment), a novel pruning technique for effectively exploring sparse structures in deep learning models, at the 34th ACM International Conference on Information and Knowledge Management (CIKM 2025) held in Seoul from 10 to 14 November.
The paper, "Exploring Diverse Sparse Network Structures via Dynamic Pruning with Weight Alignment", directly addresses the often-overlooked problem of "sparse pattern exploration" in model lightweighting. It proposes a method that fundamentally enhances the structure exploration capabilities of existing dynamic pruning techniques.
The research team stated, “DWA goes beyond merely reducing parameters; it is a method that improves the very ‘exploration capability’ of what structures sparse neural networks can adopt,” adding, “We hope it will contribute to creating lightweight yet reliable deep learning models in real-world service environments.”
This research was conducted as part of an industry-academia collaborative project supported by Hyundai Motor Company and Kia, and received funding from the AI Star Fellowship Programme (Kookmin University), supported by the Ministry of Science and ICT and the Institute for Information & Communications Technology Planning & Evaluation (IITP).
|
This content is translated from Korean to English using the AI translation service DeepL and may contain translation errors such as jargon/pronouns. If you find any, please send your feedback to kookminpr@kookmin.ac.kr so we can correct them.
|
|
New sparse neural network exploration technique “DWA” presented at CIKM 2025 / Research team led by Professor Kim Jang Ho (Department of Artificial Intelligence) |
||||
|---|---|---|---|---|
|
2025-11-25
132
Research team led by Professor Kim Jang Ho (Department of Artificial Intelligence, Kookmin University) (Kim Jin Woo, Shin Jong Yun, Ahn Sang Ho, Kim Jang Ho) presented DWA (Dynamic Pruning with Weight Alignment), a novel pruning technique for effectively exploring sparse structures in deep learning models, at the 34th ACM International Conference on Information and Knowledge Management (CIKM 2025) held in Seoul from 10 to 14 November.
The paper, "Exploring Diverse Sparse Network Structures via Dynamic Pruning with Weight Alignment", directly addresses the often-overlooked problem of "sparse pattern exploration" in model lightweighting. It proposes a method that fundamentally enhances the structure exploration capabilities of existing dynamic pruning techniques.
The research team stated, “DWA goes beyond merely reducing parameters; it is a method that improves the very ‘exploration capability’ of what structures sparse neural networks can adopt,” adding, “We hope it will contribute to creating lightweight yet reliable deep learning models in real-world service environments.”
This research was conducted as part of an industry-academia collaborative project supported by Hyundai Motor Company and Kia, and received funding from the AI Star Fellowship Programme (Kookmin University), supported by the Ministry of Science and ICT and the Institute for Information & Communications Technology Planning & Evaluation (IITP).
|
||||
