• Fan Lyu, PhD, Postdoctoral AI researcher

• New Laboratory of Pattern Recognition (NLPR)

• Institute of Automation, Chinese Academy of Sciences (CASIA)

• Contact

  : fan.lyu@cripac.ia.ac.cn, fanlyu@ieee.org

I am currently a Postdoctoral Research Fellow at the New Laboratory of Pattern Recognition (NLPR) , Institute of Automation, Chinese Academy of Sciences (CASIA) , supervised by Prof. Liang Wang . I received my Ph.D. from the College of Intelligence and Computing, Tianjin University, under the supervision of Prof. Wei Feng . Earlier, I earned my master’s degree from Suzhou University of Science and Technology in 2018, where I was advised by Prof. Fuyuan Hu . My research focuses on Open-World Learning, with an emphasis on Continual Learning and Test-Time Learning. I am particularly interested in building machine learning models that can adapt to dynamic and evolving environments. I also have a solid background in computer vision and multi-modal learning, with several publications in these domains. I have led and participated in multiple research projects and am experienced in managing teams to deliver complex tasks in both academic and applied settings.

🔔 News

• ‼️ 2025.09 | 1 paper was accepted to IEEE TCSVT 2025: MambaPTP: Exploring the Potential of Mamba for Pedestrian Trajectory Prediction. [paper]
• ‼️ 2025.09 | We achieved 4th place in the CLVISION 2025 CVIT Challenge , and our solution report will be released soon.
• ‼️ 2025.09 | 2 paper was accepted to NeurIPS 2025, and one of them is marked as spotlight:
  • (Spotlight, top 13%) Partition-Then-Adapt: Combating Prediction Bias for Reliable Multi-Modal Test-Time Adaptation
  • DAA: Amplifying Unknown Discrepancy for Test-Time Discovery
• ‼️ 2025.06 | 1 paper was accepted to IEEE TMM 2025: Constructing Enhanced Mutual Information for Online Class-Incremental Learning. [pdf] [code]
• ‼️ 2025.04 | 1 paper was accepted to IEEE TCSVT 2025: Few-Shot Class-Incremental Learning via Asymmetric Supervised Contrastive Learning. [paper]
• 2025.03 | 1 paper was accepted to ICME 2025: Controllable Continual Test-Time Adaptation. [arxiv] [code]
• 2025.02 | 3 paper was accepted to CVPR 2025:
  • Maintaining Consistent Inter-Class Topology in Continual Test-Time Adaptation [pdf] [code]
  • Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation [pdf] [code]
  • Dual Semantic Guidance for Open Vocabulary Semantic Segmentation [pdf]
• 2025.02 | 1 survey was accepted to JIG 2025: A Comprehensive Survey on Continual Learning. [paper]
• 2025.02 | Our book of continual learning is published: Continual Artificial Intelligence towards Changing Environment.
• 2025.01 | 1 paper was accepted to AAAI 2025: Rebalancing Multi-Label Class-Incremental Learning. [pdf]

📡 Research Interest

• Open-World AI: Developing AI that maintains effectiveness in dynamic and ever-changing environments
• Sustaintable AI: Ensuring AI systems remain effective and adaptive over the long term

🏹 Research Projects

• National Science Foundation of China (NSFC), 2024~2027.

• China Postdoctoral Science Foundation (CPSF), 2024~2026.

• Postdoctoral Fellowship Program of CPSF, 2023~2025.

📚 Books

Continual Artificial Intelligence towards Changing Environment