报告题目：Structured Representation and Data-Efficient Learning for Complex Biomedical Data: From Animal Behaviour Understanding to Medical Image Analysis
报告人：周飞翔
报告时间：9月14日下午18点
报告地点：理化楼405会议室
周飞翔现为英国利物浦大学博士后研究员，于2024年获得英国莱斯特大学计算机科学博士学位。在此之前，他曾任瑞德林生物技术有限公司（Readline）英国分部AI研究顾问及英国兰卡斯特大学高级研究助理。其研究兴趣主要包括机器学习、计算机视觉、视频理解和医学图像分析，重点关注面向生物医学应用的多模态学习与数据高效学习方法。其研究成果发表于多个国际权威期刊和会议，包括 IEEE Transactions on Image Processing (TIP)、IEEE Transactions on Multimedia (TMM)、IEEE Transactions on Circuits and Systems for Video Technology (TCSVT)、ECCV 和 MICCAI 等。其关于管状结构分割的研究工作入围 MICCAI 2025 Best Paper 和 Young Scientist Awards。此外，他长期担任多个国际权威期刊和顶会的审稿人，包括 IEEE TIP、TMM、TCSVT、TMI、Pattern Recognition、CVPR、ICCV、ECCV、MICCAI、ICLR、NeurIPS 和 AAAI 等。
Talk Abstract
Biomedical data are complex, structured, and heterogeneous, ranging from behavioural videos and pose trajectories to medical images and multimodal clinical records. Building reliable AI systems for such data requires addressing several recurring challenges, including complex representation learning, limited expert annotations, and incomplete or heterogeneous information. In this talk, I will present my research journey from animal behaviour understanding to medical image analysis, organised around three connected questions. First, how can complex biomedical structures be effectively represented? I will discuss work on mouse pose estimation, social interaction modelling, and structure-aware vascular segmentation. Second, how can we reduce the dependence on expensive expert annotations? I will introduce semi-supervised learning approaches for temporal action segmentation and localization, as well as semi-supervised multimodal medical image segmentation. Third, how can heterogeneous and incomplete information be effectively exploited? This includes learning shared and source-specific representations from multi-view behavioural data, multimodal medical images, clinical text and tabular data, particularly under missing-modality settings. I will conclude by discussing emerging directions in medical AI, including multimodal foundation models, AI agents for clinical workflow support, and the challenges of clinical translation.
