[Adapt] [Seminar] Rethinking classification task: label embedding and prompt learning

龚珊三 gongshansan at sjtu.edu.cn
Tue Dec 28 23:37:14 CST 2021


Hi Adapters,

I believe most of us are quite familiar with the classification task, but we usually formulate the label into the one-hot vector space and neglect its semantic meanning. In many text classification tasks, the label of input text is usually natural words, like "Politics" and "Sports", maybe we can leverage the meaning of these labels.
I will first introduce some works about label embedding, and since prompt learning is a hit topic of this year, I will also give the overview about prompt learning. At last, I will give my understanding about the “mysterious” relation between label embedding and prompt learning.

Hope you will enjoy!

Time: Wed 4:00pm
Venue: SEIEE 3-414 (I guess)
Best wishes.
Sansa


More information about the Adapt mailing list