Junkang Liu

Junkang Liu

PhD candidate in Tianjin Univeisity

刘俊康 (Junkang Liu)

刘俊康, Junkang Liu is a PhD candidate at Tianjin University, China, supervised by Prof. Fanhua Shang.
His research interests include federated learning, communication-efficient optimizers, large-model fine-tuning, model fusion, and multimodal learning.
He focuses on building efficient, secure, and scalable distributed learning systems, with an emphasis on optimization and generalization of modern large foundation models.

📧 Email: junkangliukk@gmail.com
💬 WeChat: kk15653218567
🔗 GitHub: https://github.com/junkangLiu0
🎓 Google Scholar: https://scholar.google.com/citations?hl=zh-CN&user=N7pJWIoAAAAJ —

🔥 News

  • 2026.2: 🎉🎉 Our paper ILORA were accepted by CVPR’26！
  • 2026.2: 🎉🎉 Our paper DP-FedAdamW were accepted by CVPR’26！
  • 2026.1: 🎉🎉 Our paper LA-LORA were accepted by ICLR’26！
  • 2025.10: 🎉🎉 GitHub stars have passed 600！JunkangLiu
  • 2025.10: 🎉🎉 Our paper FedAdamW were accepted by AAAI’26！
  • 2025.7: 🎉🎉 Our paper FedNSAM were accepted by ACM MM’25！
  • 2025.7: 🎉🎉 Our paper FedBCG were accepted by NeurIPS’25！
  • 2025.5: 🎉🎉 Our paper FedSWA were accepted by ICML’25！
  • 2024.7: 🎉🎉 Our paper FedBCGD were accepted by ACM MM’24！

📝 Publications

🖊️ Selected Publications ($\dagger$ denotes Corresponding Author)

  • FedAdamW: A Communication-Efficient Optimizer for Federated Large Models
    Junkang Liu, Fanhua Shang, Hongying Liu, Yuanyuan Liu, Jin Liu, Kewen Zhu, Zhouchen Lin.
    AAAI 2026 (CCF-A) [paper] [code]

  • Improving Generalization in Federated Learning via Momentum-Based Stochastic Controlled Weight Averaging
    Junkang Liu, Yuanyuan Liu, Fanhua Shang, Hongying Liu, Jin Liu, Wei Feng.
    ICML 2025 (CCF-A) [paper] [code]

  • FedBCGD: Communication-efficient Accelerated Block Coordinate Gradient Descent
    Junkang Liu, Fanhua Shang, Yuanyuan Liu, Hongying Liu, Yuangang Li, YunXiang Gong.
    ACM MM 2024 (CCF-A) [paper] [code]

  • Local-Global Flatness Consistency in Federated Learning
    Junkang Liu, Fanhua Shang, Yuxuan Tian, Hongying Liu,Yuanyuan Liu.
    ACM MM 2025 (CCF-A) [paper] [code]

  • ILoRA: Federated Learning with Low-Rank Adaptation for Heterogeneous Client Aggregation
    Junchao Zhou, Junkang Liu, Fanhua Shang.
    CVPR 2026 (CCF-A) [paper] [code]

  • DP-FedAdamW: An Efficient Optimizer for Differentially Private Federated Large Models
    Jin Liu, Ning Xi, Yinbin Miao, Junkang Liu$\dagger$.
    CVPR 2026 (CCF-A) [paper] [code]

  • Rethinking LoRA for Privacy-Preserving Federated Learning in Large Models
    Jin Liu, Ning Xi, Yinbin Miao, Junkang Liu$\dagger$.
    ICLR 2026 (CCF-A) [paper] [code]

  • High-Probability Bounds for Nonconvex Heavy-Tailed Learning
    Weixin An, Yuanyuan Liu, Fanhua Shang, Han Yu, Junkang Liu, Hongying Liu.
    NeurIPS 2025 (CCF-A) [paper]

📖 PrePrint

  • FedMuon: Accelerating Federated Learning with Matrix Orthogonalization
    Junkang Liu, Fanhua Shang, Junchao Zhou, Hongying Liu, Yuanyuan Liu, Jin Liu.
    [paper] [code]

  • Taming Preconditioner Drift: Unlocking the Potential of Second-Order Optimizers for Federated Learning on Non-IID Data
    Junkang Liu, Fanhua Shang, Hongying Liu, Jin Liu, Weixin An, Yuanyuan Liu.
    [paper] [code]

  • DP-FedPGN: Finding Global Flat Minima for Differentially Private Federated Learning via Penalizing Gradient Norm
    Junkang Liu, Yuxuan Tian, Fanhua Shang, Yuanyuan Liu, Hongying Liu, Junchao Zhou, Daorui Ding.
    [paper] [code].

  • Dynamic Differentially Private Online ADMM Algorithms with Running Average Gradients for Machine Learning
    Fanhua Shang, Junkang Liu, Weixin An, Hongying Liu.

  • IGFL:Combining Individual and Group Behaviors in Federated Learning Approaching Global Consistency
    Fanhua Shang, Junkang Liu, Weixin An, Hongying Liu.

  • LSSCA: Differentially Private Federated Learning with Laplacian Smoothing and Stochastic Controlled Averaging
    Fanhua Shang, Junkang Liu, Weixin An, Hongying Liu.

  • Towards Global Flat Minima in Sample-Level Private Federated Learning
    Jin Liu, Ning Xi, Yinbin Miao, Junkang Liu.

  • LAVA: A UNIFIED FRAMEWORK FOR FINETUNING LANGUAGE AND VISION MODELSs
    Daorui Ding, Fanhua Shang, Tiancan Feng, Junkang Liu, Hongying Liu .

🎖 Honors and Awards

  • 2024.10, National Scholarship
  • 2020.10, National Scholarship
  • 2020.10, Qingdao University Top Ten Outstanding Students Award

📖 Educations

  • 2025-, PhD in Computer Science and Technology, Tianjin University
  • 2022-2025, Master in Computer Science and Technology, Xidian University
  • 2018-2022, BSc in Mathematics, Qingdao University