[Adapt] Seminar
王绪凯
wangxukai at sjtu.edu.cn
Wed Mar 30 14:05:31 CST 2022
Hi Adapters,
This weak, I will continue my talk about methods to improve name entity recognition with distantly labeled data. Distantly labeled data have some disadvantage comparing to the human annotated ground truth data: Incomplete annotation and noisy annotation. Those disadvantages can reduce the recall and cause misprediction.
Last seminar I have introduced two papers focusing on this task. This seminar I will introduce another paper accepted by EMNLP-2021 named “Distantly-Supervised Named Entity Recognition with Noise-Robust Learning and Language Model Augmented Self-Training” They propose a noise robust learning scheme, consisting of a new loss function and a noisy label removal step. And they use language model augmented self-train to further improve the model generalization.
Hope you are interested in my talk!
Time: Wed 4:00pm
Tencent meeting number: 294-121-929
Best Wish,
Xukai Wang
More information about the Adapt
mailing list