Continual Learning Meets Real-World Visual Perception

Professor Yunchao Wei
Beijing Jiaotong University, China

Continual learning for visual perception focuses on how to continuously adapt models to new environments, identify new categories, and ultimately enhance the model's self-awareness. In this talk, Professor Wei will discuss how to conduct continual learning based on pre-trained models, how to address background drift issues in fine-grained visual perception tasks, the necessity of conducting continual learning in the era of large multimodal models, and how to empower embodied intelligence through continuous learning.

Biosketch

Yunchao Wei, Professor and Vice Dean of the School of Computer Science, Beijing Jiaotong University. He has conducted research at the National University of Singapore, the University of Illinois at Urbana-Champaign, and the University of Technology Sydney. He has been selected as MIT TR35 China, a Baidu Global High-Potential Chinese Young Scholar, and one of the "Top 40 Rising Stars" by The Australian. He has received the Pioneering Science and Technology Award at the World Internet Conference (2023), the First Prize of the Ministry of Education's Natural Science Award for Higher Education Institutions (2022), the First Prize of the China Society of Image and Graphics Science and Technology Award (2019), the Young Researcher Award from the Australian Research Council (2019), the Best Research Award from IBM C3SR (2019), the ILSVRC-Object Detection Championship (2014), and several CVPR competition championships. He has published over 100 papers in top journals/conferences such as TPAMI and CVPR, with over 20,000 citations on Google Scholar. His current research interests include visual perception for imperfect data, multimodal data analysis, and generative artificial intelligence, etc.