Xiaoyuan Zhang

Department of Computer Science, City University of Hong Kong.

prof_pic.jpg

Kowloon Tong, Hong Kong SAR China

I am a Ph.D. candidate at the Department of Computer Science, CityUHK. I will graduate in April 2025 under the supervision of Chair Prof. Zhang Qingfu (IEEE Fellow). Before joining CityUHK, I received my B.E. (2017) and M.E. (2020) degrees from Shanghai Jiao Tong University, where I was advised by Prof. Qi Chenkun. I have also been fortunate to maintain long-term collaborations with Prof. Yang Yaodong at Peking University and Prof. Han Zhao at the University of Illinois at Urbana-Champaign.

I focus on the theory, software implementation, and engineering applications of multi-objective optimization. I have made many theoretical advances in Pareto set learning and developed LibMOON, the first open-source, gradient-based library for large-scale multi-objective optimization. Collaborating with the team from PKU, I applied this technique to train large language models (LLMs) with tens of billions of parameters, with results published at NeurIPS 2024.


I will graduate in the spring of 2025 and am currently on the job market (CV, 中文简历).

Academic service

  • Conference Reviews:
    • ICLR 2025, AISTATS 2025, AAAI 2025
    • NeurIPS 2024, ICLR 2024, ICML 2024
    • ICLR 2023, NeurIPS 2023, ICML 2023
    • ICLR 2022, NeurIPS 2022, ICML 2022
    • ICML 2021
  • Journal Reviews:
    • Swarm Evolutionary Computation

Professional Talks

  1. MOO: from a Single Solution, to a Set of Solutions, and to Infinite Solutions. Institute for Artificial Intelligence, Peking University, Beijing. Host: Prof. Yang Yaodong. Mar. 8, 2024.
  2. Pareto Machine Learning: Theories, Systems, and Applications. OIST, Japan. Host: Prof. Han Zhao, Makoto Yamada. July 18, 2024.

news

Sep 26, 2024 Gliding over the Pareto Front with Uniform Designs” is accepted to NeurIPS 2024, first author.
Sep 26, 2024 Panacea: Pareto Alignment via Preference Adaptation for LLMs” is accepted to NeurIPS 2024, co-first author.
Sep 26, 2024 LibMOON: A Gradient-based MultiObjective OptimizatioN Library in PyTorch” is accepted to NeurIPS 2024, first author.
Jul 16, 2024 PMGDA: A Preference-based Multiple Gradient Descent Algorithm” is accepted to IEEE Transactions on Emerging Topics in Computing (TETCI),first author, IF=5.3.
Feb 8, 2024 Get married with Yingying Yu.
Sep 22, 2023 Hypervolume Maximization: A Geometric View of Pareto Set Learning” is accepted to NeurIPS 2023, first author.

selected publications

  1. NeurIPS
    Hypervolume maximization: A geometric view of pareto set learning
    Xiaoyuan Zhang, Xi Lin, Bo Xue, and 2 more authors
    Advances in Neural Information Processing Systems, 2023
  2. NeurIPS
    LibMOON: A Gradient-based MultiObjective OptimizatioN Library in PyTorch
    Xiaoyuan Zhang*, Liang Zhao*, Yingying Yu*, and 4 more authors
    Advances in Neural Information Processing Systems, 2024
  3. TETCI
    PMGDA: A Preference-based Multiple Gradient Descent Algorithm
    Xiaoyuan Zhang, Xi Lin, and Qingfu Zhang
    IEEE Transactions on Emerging Topics in Computational Intelligence, 2024
  4. NeurIPS
    Panacea: Pareto Alignment via Preference Adaptation for LLMs
    Yifan Zhong*, Chengdong Ma*, Xiaoyuan Zhang*, and 4 more authors
    Advances in Neural Information Processing Systems, 2024
  5. NeurIPS
    Gliding over the Pareto Front with Uniform Designs
    Xiaoyuan Zhang, Li Genghui, Xi Lin, and 3 more authors
    Advances in Neural Information Processing Systems, 2024