Abstract | ||
---|---|---|
It can be challenging to train multi-task neural networks that outperform or even match their single-task counterparts. To help address this, we propose using knowledge distillation where single-task models teach a multi-task model. We enhance this training with teacher annealing, a novel method that gradually transitions the model from distillation to supervised learning, helping the multi-task model surpass its single-task teachers. We evaluate our approach by multi-task fine-tuning BERT on the GLUE benchmark. Our method consistently improves over standard single-task and multi-task training. |
Year | DOI | Venue |
---|---|---|
2019 | 10.18653/v1/p19-1595 | ACL (1) |
DocType | Volume | Citations |
Conference | P19-1 | 2 |
PageRank | References | Authors |
0.36 | 0 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Kevin Clark | 1 | 102 | 5.93 |
Minh-Thang Luong | 2 | 1852 | 71.35 |
Urvashi Khandelwal | 3 | 275 | 10.28 |
Christopher D. Manning | 4 | 22579 | 1126.22 |
Quoc V. Le | 5 | 8501 | 366.59 |