[Adapt] [Seminar] how to leverage temporal information in our deep learning model

龚珊三 gongshansan at sjtu.edu.cn
Wed Dec 23 10:53:26 CST 2020


Hi Adapters,

As we know, in our daily life everything happens following the chronological order, and there're many expressions about time, such as timestamp, datetime, time intervals, natural language time indicators (e.g. two months ago).
In building machine learning models, time is often an important feature, and infusing temporal awareness into such models is very likely to improve the performance.

Today I will introduce some common methods and their corresponding tasks about leveraging temporal information, and I divide them into three parts: numerical representation, categorical representation and semantic meaning of time.
The first part mainly relys on rule-based modules, the second part tends to learn an embedding representation of timexes, and the last part cares about semantic meaning of time in our natural language.

Hope you will enjoy!

Time: Wed 4:00pm
Venue: SEIEE 3-414
Best wishes.
Sansa


More information about the Adapt mailing list