Chenyu Zheng (郑晨宇)   Ph.D. Student Gaoling School of Artificial Intelligence Renmin University of China Email: chenyu.zheng666[at]gmail[dot]com Links: [Google Scholar] [Github] [知乎] [CV]

About me

I am a third-year Ph.D. student in the GSAI-ML Group, Renmin University of China, advised by Prof. Chongxuan Li fortunately. Currently, I am also a research intern at ByteDance Seed. Before that, I received my B.E. degree from the School of Computer Science, Wuhan University in 2023.

Research interests

My primary interest is machine learning (theory). Recently, I focus on the optimization and scalability in deep learning.

Selected Papers

[* indicates the equal contribution, # indicates the corresponding author]

Optimization and Scalability (2023~)

• Spectral Condition for μP under Width-Depth Scaling
  Chenyu Zheng, Rongzhen Wang, Xinyu Zhang, Chongxuan Li
  Outstanding Paper Award, ICLR Workshop on Deep Generative Models: Theory, Principle, and Efficacy, 2026
  [Paper] [Code] [Blog] [Slides]
• Scaling Diffusion Transformers Efficiently via μP
  Chenyu Zheng, Xinyu Zhang, Rongzhen Wang, Wei Huang, Zhi Tian, Weilin Huang, Jun Zhu, Chongxuan Li
  Advances in Neural Information Processing Systems (NeurIPS), 2025
  [Paper] [Code] [机器之心] [Poster]
• On Mesa-Optimization in Autoregressively Trained Transformers: Emergence and Capability
  Chenyu Zheng, Wei Huang, Rongzhen Wang, Guoqiang Wu, Jun Zhu, Chongxuan Li
  Advances in Neural Information Processing Systems (NeurIPS), 2024
  [Paper] [Code] [Slides] [Poster]

Machine Learning Theory (2022~)

• Toward Understanding Generative Data Augmentation
  Chenyu Zheng, Guoqiang Wu, Chongxuan Li
  Advances in Neural Information Processing Systems (NeurIPS), 2023
  [Paper] [Code] [Slides] [Blog] [Poster]
• Revisiting Discriminative vs. Generative Classifiers: Theory and Implications
  Chenyu Zheng, Guoqiang Wu, Fan Bao, Yue Cao, Chongxuan Li, Jun Zhu
  International Conference on Machine Learning (ICML), 2023
  [Paper] [Code] [Slides] [Blog] [Poster]

Remote Sensing (2021~2022)

• Domain Adaptive Land-Cover Classification via Local Consistency and Global Diversity
  Ailong Ma, Chenyu Zheng#, Junjue Wang, Yanfei Zhong
  IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2023
  [Paper] [Code]

Experience

• Research Intern, ByteDance Seed, Beijing, China (2024.11 - Present)
  

• Research Intern, RSIDEA Group, LIESMARS, Wuhan, China (2021.01 - 2022.08)
  Advisors: Dr. Junjue Wang, Prof. Ailong Ma, and Prof. Yanfei Zhong

Selected Honors

Outstanding Paper Award, ICLR Workshop on Deep Generative Models: Theory, Principle, and Efficacy, 2026

Young Scientists (Ph.D.) Fund, National Natural Science Foundation of China, 2025

Outstanding Innovative Talents Cultivation Funded Programs, Renmin Univertity of China, 2024

Outstanding Graduate & Bachelor Thesis Award, Wuhan University, 2023

China National Scholarship, Ministry of Education of China, 2020

Talks

• Oral: Spectral Condition for μP under Width-Depth Scaling, ICLR Delta Workshop, 2026.04 [Slides]
• Tutorial: An Overview of Maximal Update Parametrization (μP), TSAIL Group@THU, 2026.04 [Slides]
• Tutorial: A Brief Review of In-Context Learning Theory, Group Session, 2024.05 [Slides]
• Tutorial: Classical Machine Learning Theory, Group Session, 2023.06 [Slides]

Academic Services

Journal Reviewer

• IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)
• IEEE Transactions on Signal Processing (TSP)

Conference Reviewer

• Advances in Neural Information Processing Systems (NeurIPS): 2024, 2025 (Top Reviewer)
• International Conference on Machine Learning (ICML): 2025, 2026
• International Conference on Learning Representations (ICLR): 2025, 2026
• International Conference on Artificial Intelligence and Statistics (AISTATS): 2025, 2026

Conference Program Committee

• Association for the Advancement of Artificial Intelligence (AAAI): 2026

Teaching

• Advanced Linear Algebra, TA, 2025 Fall
• Probability and Randomized Algorithms, TA, 2024 Fall

Miscellaneous

• Platform Sutra and meditation teach me a lot. [Platform Sutra viedo] [meditation viedo]