12.27 | Learning to Learn from Small Data
||Qianru Sun 孙倩茹
||Assistant Professor , Singapore Management University
||2019 年 12 月 27 日（周五）10 点
||张江校区软件楼 105IBM 会议室
The state-of-the-art machine learning model often requires a large number of training samples for good performance. While humans can learn new concepts and master new skills efficiently from small data. For example, kids can easily tell dogs and cats apart after seeing them only a few times, and a person who knows how to ride a bike can learn to ride a motorcycle fast with a little or even no demonstration. So, the question is "Can we design a machine learning model to have the same ability to learn fast and efficiently?" In this talk, I will introduce our meta-learning based solutions from three perspectives --- "learning to learn from pre-trained models", "learning to customize and combine multiple models", and "learning to self-train with unlabeled data", and will show concrete results accordingly.
Dr. Qianru Sun is an Assistant Professor in the School of Information Systems, Singapore Management University, since Aug 2019. Before that, she was a Joint Research Fellow working with Prof. Tat-Seng Chua at the National University of Singapore and Prof. Dr. Bernt Schiele at the MPI for Informatics (Germany) for one year. From Mar 2016 to Mar 2018, she held the Lise Meitner Award Fellowship and worked with Prof. Dr. Bernt Schiele and Dr. Mario Fritz at the MPI for Informatics. Since Jan 2016, she holds a Ph.D. degree from Peking University and her thesis was advised by Prof. Hong Liu. From Oct 2014 to Jan 2015, she was a visiting student advised by Prof. Tatsuya Harada at the University of Tokyo. Her research interests are Computer Vision and Machine Learning that aims to develop efficient learning algorithms for visual understanding.