Welcome to Peiyao Xiao's Homepage

Peiyao Xiao PhD student at UB, Computer Science and Engineering, University at Buffalo (UB) Davis Hall, Buffalo, New York Phone: +1 716-910-1694 E-mail: peiyaoxi [@] buffalo [DOT] edu

About me

I am a Ph.D. student in the Computer Science and Engineering Department at UB, supervised by Prof. Kaiyi Ji. I got the M.S. degree from King Abdullah University of Science and Technology (KAUST) in 2022. I received the B.S. degree from Southern University of Science and Technology (SUSTech).

Research

My research interests include

• Generative AI: LLM post-training (RLHF), Token pruning in VLMs

• Optimization: Multi-objective optimization, Bilevel optimization, and their applications in LLMs

• Machine learning: AI4Science, Federated learning

Find out more.

Recent Publications

• Scalable Bilevel Loss Balancing for Multi-Task Learning
  Peiyao Xiao, Chaosheng Dong, Shaofeng Zou, and Kaiyi Ji. arXiv:2502.08585.

• ReciNet: Reciprocal Space-Aware Long-Range Modeling for Crystalline Property Prediction
  Jianan Nie, Peiyao Xiao, Kaiyi Ji, Peng Gao. arXiv: 2502.02748.

• Theoretical study of conflict-avoidant multi-objective reinforcement learning
  Yudan Wang, Peiyao Xiao, Hao Ban, Kaiyi Ji, Shaofeng Zou. IEEE Transactions on Information Theory

• First-Order Federated Bilevel Learning
  Yifan Yang, Peiyao Xiao, Shiqian Ma, Kaiyi Ji. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) 2025

• MGDA Converges under Generalized Smoothness, Provably
  Qi Zhang, Peiyao Xiao, Shaofeng Zoo, and Kaiyi Ji. International Conference on Learning Representations (ICLR) 2025.

• Direction-oriented Multi-objective Learning: Simple and Provable Stochastic Algorithms
  Peiyao Xiao, Hao Ban and Kaiyi Ji. Advances in Neural Information Processing Systems (NeurIPS) 2023.

• SimFBO: Towards Simple, Flexible and Communication-efficient Federated Bilevel Learning
  Yifan Yang, Peiyao Xiao and Kaiyi Ji. Advances in Neural Information Processing Systems (NeurIPS) 2023. (Spotlight, 3% Acceptance)