[Adapt] [Seminar] Improving classification from the label space

[1840962220]

Hi Adapters,


My topic for today's seminar is: "Improving classification from the label space".


Classification is a research problem of long history in machine learning, where we want to predict discrete target variables with some input features. We have already designed many effective approaches for the input representation, including SVM, MLP, CNN, RNN, GNN, transformer, etc. However, the representation in the label space received relatively little attention, with one-hot encoding as almost the only choice. 


However, several recent works have shown to us the great potential to be explored in leveraging the rich semantics of labels. They are shown to be helpful across a variety of classification tasks, like news categorization, emotion detection and audio tagging. And I will introduce 6 representative works today.


Time: Wed 4 pm
Venue: SEIEE 3-414


Best regards,


Zhiling
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs.sjtu.edu.cn/pipermail/adapt/attachments/20210512/dcc8ecd1/attachment.html>