Qing Qu is an assistant professor in the ECE Division of the Electrical Engineering and Computer Science department of the College of Engineering, University of Michigan – Ann Arbor. He is also affiliated with the Michigan Institute for Data Science (MIDAS), the Michigan Center for Applied and Interdisciplinary Mathematics (MCAIM), and the Michigan Institute for Computational Discovery and Engineering (MICDE).

Short Bio: Dr. Qu received my B.E. degree from Tsinghua University, Beijing, China, in 2011, and obtained my Ph.D. degree from Columbia University with Prof. John Wright in 2018. He was a Moore-Sloan fellow at NYU Center for Data Science from 2018 to 2020. His work has been recognized by a couple of awards, including a Microsoft Ph.D. Fellowship in machine learning, an NSF Career Award in 2022, and an Amazon AWS AI Award in 2023. He was one of the founding organizers of the Conference on Parsimony and Learning (CPAL).

Research Interest: Broadly speaking, our research interest lies in the intersection of signal processing, data science, machine learning, and numerical optimization. In particular, I am interested in computational methods for learning low-complexity models from high-dimensional data. The current research of our group focuses on (i) the foundations of generative AI (Slides), (ii) deep representation learning (Slides-1, Slides-2), and (iii) machine learning for scientific applications.

Opening: I am always looking for self-motivated students to work with and our group typically hires 1-2 students per year, depending on the funding situation. Please fill out the form and send me an email if you are interested. Additionally, I am looking for postdocs through (i) MICDE Research Scholars, (ii) MIDAS fellow, and (iii) Schmidt AI in Science programs.

CV / Google Scholar Profile

Recent Highlights & Upcoming Events:

• Paper Acceptance: 4 papers have been accepted to NeurIPS’24 (Sept. 2024)
  • Exploring Low-Dimensional Subspaces in Diffusion Models for Controllable Image Editing
  • Understanding Generalizability of Diffusion Models Requires Rethinking the Hidden Gaussian Structure
  • BLAST: Block-Level Adaptive Structured Matrices for Efficient Deep Neural Network Inference
  • Image Reconstruction Via Autoencoding Sequential Deep Image Prior
• New Paper Release on generalizability and controllability of diffusion models (slides, presentation)
  • Diffusion Models Learn Low-Dimensional Distributions via Subspace Clustering (website)
  • Exploring Low-Dimensional Subspaces in Diffusion Models for Controllable Image Editing (website)
• Organizing a one-day MIDAS symposium on “Generative AI: From Theory to Scientific Applications” (Sept. 6th, 2024)
• Organizing a mini-symposium “Mathematical Principles in Diffusion Models” at SIAM Conference on Mathematics of Data Science (MDS24) (Oct. 21-25th, 2024)
• Organizing a session “Mathematics in Generative AI” at Asilomar Conference (Oct. 27-30th, 2024)
• Invited talk at DeepMath Conference (Nov. 14-15, 2024)

Recent Selected Publications:

• Can Yaras, Peng Wang, Laura Balzano, Qing Qu (2024). Compressible Dynamics in Deep Overparameterized Low-Rank Learning & Adaptation. International Conference on Machine Learning (ICML’24), 2024. (Oral, top 1.5%, best poster award at MMLS’24)
  Preprint – PDF –  BibTex – Code
• Huijie Zhang*, Jinfan Zhou*, Yifu Lu, Minzhe Guo, Liyue Shen, Qing Qu (2023). The Emergence of Reproducibility and Consistency in Diffusion Models. International Conference on Machine Learning (ICML’24), 2024. (Best Paper Award at NeurIPS’23 Workshop on Diffusion Models, news)
  Preprint – PDF –  BibTex – Slides – Website
• Zhihui Zhu*, Tianyu Ding*, Jinxin Zhou, Xiao Li, Chong You, Jeremias Sulam, Qing Qu (2021). A Geometric Analysis of Neural Collapse with Unconstrained Features. Neural Information Processing Systems (NeurIPS’21), 2021. (spotlight, top 3%)
  Preprint – PDF –  Slides –  BibTex – Code – Video
• Qing Qu, Yuexiang Zhai, Xiao Li, Yuqian Zhang, Zhihui Zhu (2020). Analysis of the Optimization Landscapes for Overcomplete Representation Learning. International Conference on Learning Representations (ICLR’20), 2020. (Oral, top 1.9%)
  Preprint – PDF – Slides – BibTex
• Qing Qu, Xiao Li, Zhihui Zhu (2019). Exact Recovery of Multichannel Sparse Blind Deconvolution via Gradient Descent. SIAM Journal on Imaging Science, 13(3): 1630–1652, 2020. (NeurIPS’19, spotlight, top 3%).
  Preprint – PDF – Code – Poster – Slides – BibTex
• Ju Sun, Qing Qu, John Wright (2018). A Geometric Analysis of Phase Retrieval. Foundations of Computational Mathematics, 18(5):1131–1198, 2018. (ISIT’16)
  Preprint – PDF – Code – Slides – BibTex

Funding Acknowledgement

Our group acknowledges generous support from the National Science Foundation (NSF), Office of Naval Research (ONR), Amazon Research, KLA Corporation, MICDE, and MIDAS

Contacts

Office: 4227, 1301 Beal Avenue, Ann Arbor, MI, 48109-2122