[Adapt] Seminar

[王绪凯]

Hi Adapters,
  This week, I will give you a talk about a paper accepted by ACL2021 named “Named Entity Recognition with Small Strongly Labeled and Large Weakly Labeled Data”
  Weak supervision has shown promising results in many natural language processing tasks, though still underperforms fully supervised NER. In this paper, they consider a more practical scenario, where they have both a small amount of strongly labeled data and a large amount of weakly labeled data. Unfortunately, they observe that weakly labeled data does not necessarily improve, or even deteriorate the model performance (due to the extensive noise in the weak labels). To address this issue, the authors propose a new multi-stage computational framework – NEEDLE with three essential ingredients: (1) weak label completion, (2) noise-aware loss function, and (3) final finetuning over the strongly labeled data. Through experiments on E-commerce query NER and Biomedical NER, they demonstrate that NEEDLE can effectively suppress the noise of the weak labels and outperforms existing methods.

Time: Wed 4:00pm
Venue: SEIEE 3-414(perhaps?)

Best,
Xukai Wang