Biography

I am currently a third-year Ph.D. student in the School of Computing and Data Science at The University of Hong Kong, where I am fortunate to be supervised by Prof. Yizhou Yu (ACM Fellow and IEEE Fellow).

My research focuses on building efficient foundation models for vision understanding, generation, and lifelong learning. I am particularly interested in efficient vision foundation models, continual learning, and diffusion models. My earlier work also explored cross-domain representation learning for protein language models.

Before starting my Ph.D., I received my Bachelor of Arts and Sciences degree from The University of Hong Kong in 2023, majoring in Applied Artificial Intelligence.

Research Experience

Efficient Vision Foundation Models

• Yunxiang Fu*, Meng Lou*, Yizhou Yu. SegMAN: Omni-scale Context Modeling for Semantic Segmentation, CVPR 2025 (83 citations)
  [Paper] [Code]

• Yunxiang Fu*, Chaoqi Chen*, Yizhou Yu. LaMamba-Diff: Linear-Time High-Fidelity Diffusion Models Based on Local Attention and Mamba, BMVC 2025, CORE A
  [Paper] [Code]

• Meng Lou, Yunxiang Fu, Yizhou Yu. SparX: Sparse Cross-Layer Connection for Vision Mamba/Transformer, AAAI 2025, CCF-A
  [Paper] [Code]

Continual Learning for Vision Foundation Models

• Yunxiang Fu, Meng Lou, Yizhou Yu. One Adapter, Many Tasks: Task-Conditioned Feature Transformations for Continual Learning, NeurIPS 2026 submission

• Meng Lou, Yunxiang Fu, Yizhou Yu. Scaling Continual Learning to 300+ Tasks with Bi-Level Routing MoE, ICML 2026, CCF-A
  [Paper] [Code]

Diffusion Models

• Yunxiang Fu, Chaoqi Chen, Yu Qiao, Yizhou Yu. DreamDA: Generative Data Augmentation with Diffusion Models, journal submission
  [Paper] [Code]

Protein Language Models

• Hong-Yu Zhou*, Yunxiang Fu*, Zhicheng Zhang, Cheng Bian, Yizhou Yu. Protein Representation Learning via Knowledge Enhanced Reasoning, ICLR 2023, CCF-A (63 citations)
  [Paper] [Code]

For the full and most up-to-date publication list, please see my Google Scholar profile.

News

• 2026.05: Our single-adapter continual learning work is submitted to NeurIPS 2026.
• 2026.02: CaRE, our work on scaling continual learning to 300+ tasks with bi-level routing mixture-of-experts, is available on arXiv.
• 2025.07: LaMamba-Diff is accepted to BMVC 2025.
• 2025.05: DreamDA is submitted as a journal manuscript.
• 2025.02: SegMAN is accepted to CVPR 2025.
• 2024.09: SparX is accepted to AAAI 2025.

Academic Service

I serve as a reviewer for journals and conferences including TPAMI, TIP, CVPR, ECCV, AAAI, ICML, ICLR, and NeurIPS.