Guojun Zhang

I am an algorithmic engineer at MiniMax. My current research focus is on Alignment for LLMs, and transfer learning. I obtained my Ph.D [thesis] from the David R. Cheriton School of Computer Science at the University of Waterloo, also as a student affiliate of the Vector Institute. My PhD supervisors were Prof. Pascal Poupart and Prof. Yaoliang Yu. I obtained my master in theoretical physics from the Perimeter Institute, and I was fortunate to work with Prof. Freddy Cachazo on theoretical physics.

Research Interests: alignment, transfer learning, multimodality

News

• 2024.07 I started my new job at MiniMax. Excited to ramp up in LLMs.
• 2024.06 One paper accepted to ICML 2024.
• 2023.12 Our paper ''Calibrated One Round Federated Learning with Bayesian Inference in the Predictive Space'' has been accepted in AAAI 2024. Congrats Mohsin (now a PhD student of Yoshua Bengio)!
• 2023.08 I will serve as an Area Chair for AISTATS 2024.
• 2023.08 Our paper ''Private GANs, Revisited'' has been accepted to TPDP 2023 and TMLR.
• 2023.07 Our paper ''Understanding Hessian Alignment for Domain Generalization'' has been accepted to ICCV 2023.
• 2022.12 Our paper ''Proportional Fairness in Federated Learning'' has been accepted at TMLR.
• 2022.11 Two papers to present at NeurIPS 2022: our JMLR paper at the Journal-to-Conference Track and our paper ''Private GANs, Revisited'' at the SyntheticData4ML Workshop.
• 2022.11 Reviewing for ICLR 2023.
• 2022.10 Reviewing for SyntheticData4ML Workshop in NeurIPS 2022.
• 2022.08 I will serve as an Area Chair for AISTATS 2023.
• 2022.07 I will serve as a Session Chair in ICML 2022.
• 2022.05 I will give an invited talk at 2022 Optimization Days organized by HEC Montréal.
• 2022.04 Our paper ''Federated Learning Meets Multi-objective Optimization'' is accepted at IEEE Transactions on Network Science and Engineering!
• 2022.03 I will serve as a Program Committee in the FL-IJCAI workshop 2022.
• 2022.01 Our paper ''Domain Adversarial Training: A Game Perspective'' has been accepted at ICLR 2022.
• 2022.01 Our paper ''Optimality and Stability in Non-convex Smooth Games'' has been accepted to Journal of Machine Learning Research.
• 2021.11 Our paper ''f-Mutual Information Contrastive Learning'' has been accepted as an oral to NeurIPS 2021 workshop on self-supervised learning!
• 2021.09 Our paper ''Quantifying and Improving Transferability in Domain Generalization'' is accepted at NeurIPS 2021! Thanks to all my collaborators!
• 2021.09 Excited to start my new job as a senior researcher at Huawei Montreal!
• 2021.07 Our paper ''Newton-type Methods for Minimax Optimization'' is accepted at the ICML 2021 workshop ''Beyond first-order methods in ML systems''
• 2021.07 Completed my defence! Thanks to all committee members!
• 2021.05 Our paper ''f-Domain Adversarial Learning: Theory and Algorithms'' is accepted at ICML 2021
• 2021.05 Happy to serve as reviewers for CoRL 2021 and NeurIPS 2021
• 2020.08 Attending CIFAR Deep Learning + Reinforcement Learning Summer School
• 2020.05 I'm excited to join Prof. Sanja Fidler's group to do a summer intern at NVIDIA, Toronto.
• 2020.04 Attending ICLR 2020 and presenting my work on bilinear zero-sum games

Education

• 2017.09-2021.08 Ph.D. of Computer Science, University of Waterloo, Waterloo, Canada. Supervisors: Pascal Poupart and Yaoliang Yu
• 2015.08-2016.06 Master of Physics, University of Waterloo/Perimeter Institute, Waterloo, Canada. Supervisor: Freddy Cachazo
• 2011.09-2015.07 Bachelor of Physics, University of Science and Technology of China, Hefei, China. GPA—4.13/4.3

Awards

• ICML 2022 Top 10% Reviewer
• Cheriton Scholarship
• Guo Moruo Scholarship

Selected Publications

Transfer Learning (Domain Generalization/Adaptation/Multitask):
• 2023.05 Yifei He, Shiji Zhou, Guojun Zhang, Hyokun Yun, Yi Xu, Belinda Zeng, Trishul Chilimbi, Han Zhao. Robust Multi-Task Learning with Excess Risks. ICML 2024.
  [ICML][arxiv]
• 2023.11 (OOD) Ahmad Rashid, Serena Hacker, Guojun Zhang, Agustinus Kristiadi, Pascal Poupart. Preventing Arbitrarily High Confidence on Far-Away Data in Point-Estimated Discriminative Neural Networks. AISTATS 2024.
  [AISTATS][arxiv]
• 2023.07 (DG) Sobhan Hemati*, Guojun Zhang*, Amir Estiri, Xi Chen. Understanding Hessian Alignment for Domain Generalization . ICCV 2023.
  [ICCV][arxiv][code][poster][video]
• 2022.11 (DA) Ehsan Imani, Guojun Zhang, Jun Luo, Pascal Poupart, Philip Torr, Yangchen Pan. Label Alignment Regularization for Distribution Shift. JMLR 2024. arXiv preprint arXiv:2211.14960.
  [arxiv]
• 2022.01 (DA) David Acnua, Marc Law, Guojun Zhang, Sanja Fidler. Domain Adversarial Training: A Game Perspective. ICLR 2022.
  [ICLR][arxiv]
• 2021.06 (DG) Guojun Zhang, Han Zhao, Yaoliang Yu and Pascal Poupart. Quantifying and Improving Transferability in Domain Generalization. NeurIPS 2021.
  [NeurIPS][arxiv][code][video]
• 2020.10 (DA) David Acuna, Guojun Zhang, Marc Law and Sanja Fidler. f-Domain Adversarial Learning: Theory and Algorithms. ICML 2021 (spotlight).
  [ICML][arxiv][code]
  
  Federated learning and privacy:
• 2024.08 Yasser H. Khalil, Amir Hossein Estiri, Mahdi Beitollahi, Nader Asadi, Sobhan Hemati, Xu Li, Guojun Zhang, Xi Chen. DFML: Decentralized Federated Mutual Learning. TMLR 2024.
[TMLR][arxiv]
• 2023.12 Mohsin Hasan, Guojun Zhang, Kaiyang Guo, Xi Chen, Pascal Poupart. Calibrated One Round Federated Learning with Bayesian Inference in the Predictive Space. AAAI 2024.
[AAAI][arxiv][code]
• 2023.08 Guojun Zhang, Mahdi Beitollahi, Alex Bie, Xi Chen. Understanding the Role of Layer Normalization in Label-Skewed Federated Learning. TMLR 2024.
  [TMLR][arxiv][code]
• 2022.11 Alex Bie, Gautam Kamath, Guojun Zhang. Private GANs, Revisited. TMLR 2023 ; NeurIPS 2022 SyntheticData4ML workshop; TPDP 2023.
  [TMLR][NeurIPS workshop][arxiv]
• 2022.06 Artur Back de Luca*, Guojun Zhang*, Xi Chen, Yaoliang Yu. Mitigating Data Heterogeneity in Federated Learning with Data Augmentation.
  [arxiv][code]
• 2022.02 Guojun Zhang, Saber Malekmohammadi, Xi Chen and Yaoliang Yu. Proportional Fairness in Federated Learning. TMLR 2023.
  [TMLR][arxiv][slides][code]
• 2020.06 Zeou Hu, Kiarash Shaloudegi, Guojun Zhang and Yaoliang Yu. Federated Learning meets Multi-objective Optimization. IEEE Transactions on Network Science and Engineering 2022.
  [IEEE TNSE][arxiv][code]
  
  Contrastive Learning:
• 2021.12 Yiwei Lu*, Guojun Zhang*, Sun Sun, Hongyu Guo and Yaoliang Yu. $f$-MICL: Understanding and Generalizing InfoNCE-based Contrastive Learning. TMLR 2023. NeurIPS 2021 workshop on self-supervised learning (contributed talk).
  [TMLR][NeurIPS workshop][poster]
  
  Minimax Optimization and Smooth Games:
• 2022.01 Guojun Zhang, Pascal Poupart and Yaoliang Yu. Optimality and Stability in Non-Convex Smooth Games. JMLR 2022.
  [JMLR][arxiv][bib]
• 2020.06 Guojun Zhang, Kaiwen Wu, Pascal Poupart and Yaoliang Yu. Newton-type Methods for Minimax Optimization. ICML 2021 workshop for ''Beyond first-order methods in ML systems.''
  [arxiv][workshop][code]
• 2019.08 Guojun Zhang and Yaoliang Yu. Convergence of Gradient Methods on Bilinear Zero-Sum Games.
  ICLR 2020, also presented at NeurIPS workshop SGO&ML 2019
  [ICLR][arxiv][workshop][code]
  
  Theoretical Physics:
• 2017.05 Sebastian Mizera and Guojun Zhang (α-β order). String-theoretical Deformation of the Parke-Taylor Factor. Phys. Rev. D 96 (2017) no.6, 066016.
  [PRD][arxiv]
• 2016.12 Humberto Gomez, Sebastian Mizera and Guojun Zhang (α-β order). CHY Loop Integrands from Holomorphic Forms. JHEP 1703 (2017) 092.
  [JHEP][arxiv]
• 2016.09 Freddy Cachazo, Sebastian Mizera and Guojun Zhang (α-β order). Scattering Equations: Real Solutions and Particles on a Line. JHEP 1703 (2017) 151.
  [JHEP][arxiv]
• 2015.05 Xin Wang, Guojun Zhang and Min-xin Huang, New Exact Quantization Condition for Toric Calabi-Yau Geometries. Phys. Rev. Lett. 115, 121601 (2015).
  [PRL][arxiv]

Academic Services

PC Member: IJCAI
Conference Reviewer: NeurIPS, ICML, ICLR, AISTATS, CoRL, AAAI
Program Committee: FL-IJCAI 2022, FL@FM-IJCAI 2024, FL@FM-ICME 2024
Journal Reviewer: Journal of Scientific Computing, TMLR, ACM Transactions on Intelligent Systems and Technology
Workshop Reviewer: SyntheticData4ML Workshop NeurIPS 2022.

Misc

• General: How to read a paper; How to write an ML paper; How to write math; How to give a talk
• Math classics:
  • Analysis: Baby Rudin; Child Rudin (on it now); Adult Rudin
  • Probability and statistics: Stochastic Processes; Dembo's note
  • Matrices: Random matrix theory; Matrix analysis
  • Algebra: Invariant theory
  • (to be continued, suggestions are welcome)