Zicong Zhang

latest update: Nov 2, 2026

Hi! I’m Zicong Zhang, an undergraduate student majoring in Computer Science in the John Hopcroft Class at Zhiyuan College, Shanghai Jiao Tong University. Since August 2026, I have been a visiting student at Harvard Vision and Learning Lab, working with Prof. Qianqian Wang.

Previously, I was a research intern at Microsoft Research Asia, advised by Yifan Yang and Chong Luo, and at SJTU MIFA Lab, advised by Prof. Weiran Huang.


I’m interested in building Versatile, Intelligent, Ever-Learning Agents that continuously learn and reason across vision, language, and real-world environments. My current research spans:

  • Generative Modeling & Post-Training: distribution matching and Jacobian-aware post-training for efficient one-step generation.
  • Self-Evolving Agents & Environment Optimization: closing the loop between behavioral diagnosis, environment construction, reinforcement learning, and evaluation.
  • Multimodal Reasoning & Continual Learning: integrating reasoning with visual tools and enabling vision-language models to acquire and retain new knowledge.

news

Oct 2026 We have completed EnvOpt: Closed-Loop Environment Optimization for Agent Post-Training! The detailed framework and code will be released soon. Stay tuned!
Aug 2026 I’m fortunate to join Harvard Vision and Learning Lab as a visiting student, working with Prof. Qianqian Wang!
Jan 26, 2026 Co-first author paper Enhanced Continual Learning of Vision-Language Models With Model Fusion has been accepted by 14th ICLR! See U in :brazil: ~
Jul 17, 2025 I’m fortunate to join Microsoft Research Asia as a research intern, advised by Yifan Yang !
Mar 06, 2025 Co-first author paper Enhanced Continual Learning of Vision-Language Models With Model Fusion has been accepted by 13th ICLR SCOPE Workshop ! :sparkles:

selected publications

  1. ConDU.png
    Enhanced Continual Learning of Vision-Language Models with Model Fusion
    Haoyuan Gao*, Zicong Zhang*, Yuqi Wei, and 5 more authors
    In The Fourteenth International Conference on Learning Representations, 2026
  2. XSearch: Explainable Code Search via Concept-to-Code Alignment
    Yiming Liu, Ruofan Liu, Yun Lin, and 7 more authors
    In The 35th ACM SIGSOFT International Symposium on Software Testing and Analysis (ISSTA), 2026
  3. EnvOpt: Closed-Loop Environment Optimization for Agent Post-Training
    Zicong Zhang*, Zisu Huang*, Jingwen Xu, and 10 more authors
    2026
    Under Review