基于轻量化迁移学习的云边协同自然语言处理方法
|
赵蕴龙,赵敏喆,朱文强,查星宇
|
Cloud-edge collaborative natural language processing method based on lightweight transfer learning
|
Yunlong ZHAO,Minzhe ZHAO,Wenqiang ZHU,Xingyu CHA
|
|
表 2 本研究所提方法与基线模型在4个公开数据集上的实验结果 |
Tab.2 Experimental results of proposed method compared with baseline model on four public datasets |
|
方法 | A(σ) | M(σ) | F(σ) | P(σ) | BERT Fine-tuning | 61.85(2.92) | 55.21(0.94) | 86.97(2.85) | 88.27(0.17) | Adapter | 61.97(3.11) | 55.14(0.39) | 85.21(0.39) | 88.51(0.15) | Prefix-tuning | 58.84(0.36) | 38.91(0.51) | 76.59(0.36) | 79.16(0.24) | LoRA | 66.06(1.25) | 56.24(1.94) | 86.49(0.30) | 88.61(0.16) | Cloud-Edge Collaborative Model (本方法) | 60.17(1.03) | 48.06(1.26) | 80.62(0.40) | 87.48(0.16) | w/o the text embedding module | 58.21(0.95) | 46.13(0.63) | 78.64(0.31) | 84.26(0.09) | w/o the task transfer module | 58.86(0.22) | 47.14(1.24) | 78.87(0.50) | 85.78(0.94) |
|
|
|