low-rank approximation 相关 AI 资讯

11:15

官方账号arXiv cs.LG@Hao Luo, Yiting Yang, Wenyi Zhao, Man Jiang, Zhijun Lin, Ghulam Mohiuddin, Ting Jiang, Kunming Luo, Zihao Zhang, Qingsen Yan, Guoqing Wang, Wei Dong, Peng Wang

We propose Channel Group-Shared (CGS) low-rank approximation for efficient large-kernel CNNs, reducing parameter volume by 87% in models like RepLKNet-31B. CGS achieves significant parameter reduction and enables feasible deployment on edge devices, improving performance and reducing storage costs.

推荐理由：Read this if you're interested in how CGS low-rank approximation can make large-kernel CNNs more efficient for mobile devices, compared to existing methods like RepLKNet-31B.

原文