About Me
Hi! I’m Bohang Zhang (张博航), a Ph.D. graduate at Peking University, advised by Prof. Liwei Wang. I also work closely with Prof. Di He. Before starting Ph.D., I completed my undergraduate studies at School of the Gifted Young (少年班) in Xi’an Jiaotong University, majoring in Computer Science.
My main research area lies in studying the foundations of machine learning, such as the expressive power, robustness, and optimization of neural networks. My work provides theoretical and algorithmic insights into the strengths and weaknesses of fundamental deep learning models and algorithms (often through a computer science perspective), based on which I design new (provably better) models/algorithms. My first-authored papers have won the ICLR 2023 Outstanding Paper Award and ICLR 2024 Outstanding Paper Honorable Mention. My Ph.D. thesis has been selected as the Outstanding Doctoral Dissertation in Peking University.
Here are several research areas I worked on:
- Understanding the power and limitations of large language models (LLMs) in complex reasoning.
- Analyzing the expressive power of graph neural networks (GNNs), providing guaidance on the GNNs design principles that enable them to effectively represent the necessary graph structural information.
- Designing powerful Lipschitz neural networks with certified robustness gurantees, i.e. achieving provable robustness under adversarial attacks.
- Designing and analyzing optimization algorithms for efficient neural network training.
If you are interested in collaborating with me or want to have a chat, always feel free to contact me through e-mail or Wechat.
📝 Publications
* means equal contribution. See the Publications page for more details.
- On the Expressive Power of Spectral Invariant Graph Neural Networks.
Bohang Zhang, Lingxiao Zhao, Haggai Maron. In ICML 2024. [Code] - Do Efficient Transformers Really Save Computation.
Kai Yang, Jan Ackermann, Zhenyu He, Guhao Feng, Bohang Zhang, Yunzhen Feng, Qiwei Ye, Di He, Liwei Wang. In ICML 2024. - Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness.
Bohang Zhang*, Jingchu Gai*, Yiheng Du, Qiwei Ye, Di He, Liwei Wang. In ICLR 2024. [Code]
(Outstanding Paper Award Honorable Mention, 0.23% acceptance rate) - Towards Revealing the Mystery behind Chain of Thought: A Theoretical Perspective.
Guhao Feng*, Bohang Zhang*, Yuntian Gu*, Haotian Ye*, Di He, Liwei Wang. In NeurIPS 2023. [Code]
(Oral Presentation, 0.6% acceptance rate) - A Complete Expressiveness Hierarchy for Subgraph GNNs via Subgraph Weisfeiler-Lehman Tests.
Bohang Zhang, Guhao Feng*, Yiheng Du*, Di He, Liwei Wang. In ICML 2023. [Code]
(All ratings are clear acceptance) - Rethinking the Expressive Power of GNNs via Graph Biconnectivity.
Bohang Zhang*, Shengjie Luo*, Liwei Wang, Di He. In ICLR 2023. [Code]
(Outstanding Paper Award, top 4/4966) - Finding Generalization Measures by Contrasting Signal and Noise.
Jiaye Teng*, Bohang Zhang*, Ruichen Li*, Haowei He*, Yequan Wang, Yan Tian, Yang Yuan. In ICML 2023. - Rethinking Lipschitz Neural Networks and Certified Robustness: A Boolean Function Perspective.
Bohang Zhang, Du Jiang, Di He, Liwei Wang. In NeurIPS 2022. [Code]
(Oral Presentation, 1.7% acceptance rate) - Boosting the Certified Robustness of L-infinity Distance Nets.
Bohang Zhang, Du Jiang, Di He, Liwei Wang. In ICLR 2022. [Code] - Towards Certifying L-infinity Robustness using Neural Networks with L-inf-dist Neurons.
Bohang Zhang, Tianle Cai, Zhou Lu, Di He, Liwei Wang. In ICML 2021 (Spotlight). [Code] - Non-convex Distributionally Robust Optimization: Non-asymptotic Analysis.
Jikai Jin*, Bohang Zhang*, Haiyang Wang, Liwei Wang. In NeurIPS 2021. - Improved Analysis of Clipping Algorithms for Non-convex Optimization.
Bohang Zhang*, Jikai Jin*, Cong Fang, Liwei Wang. In NeurIPS 2020. [Code]
🎖 Selected Awards
- ICLR 2023 Outstanding Paper Award (top 4/4966). [Link] [Certificate]
- ICLR 2024 Outstanding Paper Award Honorable Mention (top 17/7262). [Link] [Certificate]
- Outstanding Doctoral Dissertation. Awarded annually to about 3% Ph.D. graduates in Peking University.
- Bytedance Scholarship, 2023. Awarded annually to about 10 students in China. [Link] [Certificate]
- Principal Scholarship, 2019-2020, 2020-2021, 2021-2022, 2022-2023, 2023-2024. Awarded annually to top 1 student in the same grade in School of Artificial Intelligence, Peking University. [Certificate] [Certificate] [Certificate] [Certificate] [Certificate]
- ACM ICPC World Finalist (ranking 41/135), Porto Portugal, 2019. [Certificate] [Certificate]
- Beijing outstanding PhD graduates, 2024.
- ACM ICPC East Asia Continent Final Gold Award (ranking 8/382), Xi’an China, 2018. [Certificate][Certificate]
- ACM ICPC 2nd Runner up (Gold Award, ranking 4/298), Jiaozuo China, 2018. [Certificate]
- Top-10 outstanding student pioneers (ranking 2/10), 2019. Awarded annually to a total of 10 undergraduate students across Xi’an Jiaotong University.
💬 Invited Talks
- Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness.
- 2023.6.27. Hosted by AI Time. [News] [Slides]
- 2023.7.9. Hosted by Prof. Christopher Morris and Luis Müller in RWTH Aachen University. [Slides]
- 2023.7.19. Hosted by FAI Seminar. [News] [Slides]
- 2023.7.21. Hosted by CCF Theoretical Computer Science Symposium. [Handbook] [Slides]
- Towards Revealing the Mystery behind Chain of Thought: A Theoretical Perspective.
- 2023.6.8. Huawei Noah’s Ark Lab. [Slides]
- 2023.11.17. Hosted by FAI Seminar. [Slides]
- 2023.11.20. Hosted by FLaNN seminars. [News] [Video] [Slides]
- 2024.6.2. Hosted by CCF Theoretical Computer Science Committee Outstanding Doctoral Student Forum. [Poster] [Slides]
- Understanding and Improving the Expressivity of Subgraph GNNs.
- 2023.6.23. Hosted by FAI Seminar. [News] [Slides] [Video]
- Rethinking the Expressive Power of GNNs via Graph Biconnectivity.
- 2023.3.31. Hosted by FAI Seminar and Jiangmen TechBeat. [News] [Poster] [Slides] [Video]
- 2023.4.9. Hosted by Learning on Graph Seminar. [News] [Slides] [Video]
- 2023.5.26. Hosted by CMLR at Peking University. [Poster] [Slides]
- 2023.6.7. Hosted by WestLake University. [Poster] [Slides]
- 2023.6.12. Hosted by VALSE. [Agenda] [Slides]
- Understanding and Improving Expressive Power of GNNs: Distance, Biconnectivity, and WL Tests.
- 2023.3.16. Hosted by Prof. Haggai Maron at Technion. [Slides]
- Rethinking Lipschitz Neural Networks and Certified Robustness: A Boolean Function Perspective.
- 2022.12.15. Hosted by Qiongxiu Li at Tsinghua University. [Poster] [Slides]
- 2022.11.10. Hosted by Prof. Yong Liu at Remin University of China. [Slides]
- 2022.12.21. Hosted by CVMart (极市平台). [News] [Poster] [Video] [Slides]
- 2022.11.26. 2022 NeurIPS Meetup China by Synced (机器之心). [News] [Poster] [Slides]
- Non-convex Distributionally Robust Optimization: Non-asymptotic Analysis.
- 2022.3.10. Huawei Noah’s Ark Lab. [Slides]
- Analyzing and Understanding Gradient Clipping in Non-Convex Optimization.
🏫 Professional Services
- Workshop Organizer for ICLR 2024.
- Reviewer for ICML 2022, NeurIPS 2022 (top reviewer), ICLR 2023, CVPR 2023, ICML 2023, VFVML 2023, NeurIPS 2023 (top reviewer), ICLR 2024, ICML 2024.
- Reviewer for Transactions on Pattern Analysis and Machine Intelligence (TPAMI).