报告题目Title:
Label and Sample Efficient Learning for Semantic Segmentation and Anomaly Detection
主讲人Speaker：
肖继民教授，西交利物浦大学
Prof. Jimin Xiao, Xi’an Jiaotong-Liverpool University
日期和时间DATE & TIME:
2026年2月4日，10:00-11:00
Wednesday February 4, 2026, 10:00-11:00am
地点VENUE:
办公楼106，永久校区
Room 106, Office Building, Permanent Campus
腾讯会议Tencent Meeting ID：
528-894-371
报告摘要Abstract:
This talk focuses on label- and sample-efficient learning for semantic segmentation and industrial anomaly detection. Regarding semantic segmentation, it covers weakly supervised semantic segmentation, where annotations are at the image or bounding box level—rather than the costly pixel-level labels. Another approach to reducing annotation costs is semi-supervised learning, in which only a subset of training samples is equipped with pixel-level labels, while the rest are unlabeled. For few-shot segmentation, only a small number of annotated samples are required for new classes; this method leverages meta-learning to transfer segmentation capabilities from a large set of annotated samples belonging to other classes. We also extend our research to unsupervised and few-shot anomaly detection, which utilizes no or very limited defect samples and annotations
主讲人介绍Biography：
肖继民教授于2013年获得英国利物浦大学博士学位，2013-2014年在芬兰诺基亚研究院担任高级研究员。2014年起在西交利物浦大学智能工程学院任职，担任英国利物浦大学博士生导师，已指导毕业博士生10多人。毕业的多名博士生已经以副教授身份入职苏州大学、中国石油大学和上海人工智能研究院等知名高校和研究所。
肖继民教授已主持国家自然科学基金面上项目（2项）、青年项目和重点项目课题。于2016年获评苏州市高等学校科研院所紧缺人才，于2022年获评江苏省333高层次人才。其学术论文发表于人工智能领域各大顶级期刊和会议，在IEEE TPAMI, IJCV, NeurIPS, CVPR, ICCV, ECCV, AAAI, IJCAI等期刊和会议发表论文100多篇。其中IEEE Transactions和CCF-A类会议论文60多篇。主要研究方向包括计算机视觉，人工智能。
主持人Host：
梁杰
信息学部讲席教授
加拿大工程院院士
Prof. Jie Liang
Chair Professor, EE School, EIT
Fellow, Canadian Academy of Engineering
