新加坡科研技术局（Agency for Science, Technology and Research，A*STAR）高级经理（Senior Manager）Dr. Yiqun Li 报告会
Topic：Incremental Learning with A Few Examples for industry applications
Time：2019年9月24日（周二）上午9:30 – 11:00
Venue：玉泉校区教一103会议室
Speaker：Dr. Yiqun Li, Institute for Infocomm Research (I2R), Agency for Science, Technology and Research (A*STAR), Singapore
Incremental Learning with A Few Examples forindustry applications
There are many industry applications require incremental learning using a few examples. However, this is a challenging problem as standard Deep Neural Networks (DNNs) is not able to deal with it due to the two problems below:
(1) DNNs suffer from catastrophic forgetting when used for incremental class learning;
(2) DNNs do not perform well when little training data is available.
In this talk, we introduce MetaMG - Meta Module Generation to address above two problems simultaneously. MetaMG is a meta-learning method that enables a module generator to rapidly generate a category module from a few examples for a scalable classification network to recognize a new category. The old categories are not forgotten after new categories are added in. We will show experimental results on a few datasets for discussion.
Short Bio:
Yiqun Li is a Senior Manager at the Institute for Infocomm Research (I2R), A*STAR with over 20 years of R&D experience on computer vision technologies in both research institute and MNC. She is also the Assistant Head of the Advanced Manufacturing Engineering (AME) Division and the Unit Head of Defect Detection and Object Recognition Unit. Her current research interests are focused on image classification, object detection and segmentation using convolutional neural networks, incremental learning, meta learning and few-shot learning, with applications on defection detection, measurement and classification in AME domain. She has been the Principal Investigator for many research projects in the areas of place recognition, augmented reality, object instance recognition, image and video analysis etc. Many of her research outcomes were successfully translated into industry applications. She has 8 patents and a number of scientific publications.