报告人： 吴敏 研究科学家 新加坡科技研究局（A*STAR）
报告题目：Label-efficient Time Series Representation Learning
时间： 2023年6月5日 (周一) 上午10:00 -11:30
地 点： 信息科技大楼1715会议室（线下）
腾讯会议号：755-792-886（线上）
主持：徐 军 教 授
报告摘要:
The scarcity of labeled data is one of the main challenges of applying deep learning models on time series data in the real world. Therefore, several approaches, e.g., transfer learning, self-supervised learning, and semi-supervised learning, have been recently developed to promote the capability of deep learning models from the limited time series labels. In this talk, we will briefly review the existing approaches that address the scarcity of labeled data problem in time series data, and categorize them as in-domain solutions and cross-domain solutions. Then, we will share our own capabilities in this topic. We first introduce two self-supervised learning methods called TS-TCC and CA-TCC for in-domain time series representation learning. We also present our method called SEA and our benchmarking suite called AdaTime as cross-domain solutions. Lastly, we discuss the limitations of these methods and provide future directions in this field.
报告人简介:吴敏博士目前担任新加坡科技研究局（A*STAR）机器智能部门担任资深研究科学家。研究领域包括时间序列数据和图数据的机器学习和数据挖掘。分别于2011年在新加坡南洋理工大学 (NTU) 获得计算机科学博士学位和2006年在中国科学技术大学(USTC)获得计算机科学学士学位。曾获得了2022 IEEE ICIEA、2022 IEEE SmartCity、2016 InCoB和2015 DASFAA的最佳论文奖，以及2020 IEEE PHM的入围学术论文奖，以及分别获得了2021 CVPR UG2+ 挑战赛和 2015 年 IJCAI competition on repeated buyers prediction的冠军。
