Electrical and Computer Engineering · University of Michigan, Ann Arbor
PI: Prof. Qing Qu
Our research lies at the intersection of signal processing, data science, machine learning, and numerical optimization. We are interested in computational methods for learning low-complexity models from high-dimensional data. Current directions:
- Foundations of generative AI
- Deep representation learning
- Machine learning for scientific applications
Code released with our publications. Click a title to open the repository; each one is forked here from the authors' original repository.
2026 (4)
Breaking the Curse of Dimensionality: Diffusion Models Efficiently Learn Low-Dimensional Distributions
JMLR 2026 · Paper · Project
DCDP: Decoupled Data Consistency with Diffusion Purification for Image Restoration
ICCP 2026 · Paper
Generalization of Diffusion Models Arises with a Balanced Representation Space
ICLR'26 · Oral at DeepMath'25 · Paper · Project
AlphaFlow: Understanding and Improving MeanFlow Models
ICLR'26 · Paper
2025 (3)
A Closer Look at Model Collapse: From a Generalization-to-Memorization Perspective
NeurIPS'25 · Spotlight (top 3.2%) · Paper · Project
Shallow Diffuse: Robust and Invisible Watermarking through Low-Dimensional Subspaces in Diffusion Models
NeurIPS'25 · Spotlight (top 3.2%) · Paper · Project
Understanding Deep Representation Learning via Layerwise Feature Compression and Discrimination
JMLR 2025 · Paper
2024 (5)
BLAST: Block-Level Adaptive Structured Matrices for Efficient Deep Neural Network Inference
NeurIPS'24 · Paper
Image Reconstruction Via Autoencoding Sequential Deep Image Prior
NeurIPS'24 · Paper
Exploring Low-Dimensional Subspaces in Diffusion Models for Controllable Image Editing
NeurIPS'24 · Paper · Project
Compressible Dynamics in Deep Overparameterized Low-Rank Learning & Adaptation
ICML'24 · Oral (top 1.5%) · Paper
Optimal Eye Surgeon: Finding Image Priors Through Sparse Generators at Initialization
ICML'24 · Paper
2023 (4)
Improving Efficiency of Diffusion Models via Multi-Stage Framework and Tailored Multi-Decoder Architectures
CVPR'24 · Paper · Project
Efficient Low-Dimensional Compression of Overparameterized Models
AISTATS'24 · Paper
Neural Collapse in Multi-label Learning with Pick-all-label Loss
ICML'24 · Paper
The Law of Parsimony in Gradient Descent for Learning Deep Linear Networks
arXiv 2023 · Paper
2022 (5)
Neural Collapse with Normalized Features: A Geometric Analysis over the Riemannian Manifold
NeurIPS'22 · Paper
Are All Losses Created Equal: A Neural Collapse Perspective
NeurIPS'22 · Paper
Hidden State Variability of Pretrained Language Models Can Guide Computation Reduction for Transfer Learning
Findings of EMNLP 2022 · Paper
Robust Training under Label Noise by Over-parameterization
ICML'22 · Paper
On the Optimization Landscape of Neural Collapse under MSE Loss: Global Optimality with Unconstrained Features
ICML'22 · Paper
2021 (3)
A Geometric Analysis of Neural Collapse with Unconstrained Features
NeurIPS'21 · Spotlight (top 3%) · Paper
Convolutional Normalization: Improving Deep Convolutional Network Robustness and Training
NeurIPS'21 · Paper
Weakly Convex Optimization over Stiefel Manifold Using Riemannian Subgradient-Type Methods
SIAM J. Optim. 2021 · Paper
2020 (2)
Robust Recovery via Implicit Bias of Discrepant Learning Rates for Double Over-parameterization
NeurIPS'20 · Spotlight (top 4%) · Paper
Short and Sparse Deconvolution — A Geometric Approach
ICLR'20 · Paper · Project
2019 & Prior (6)
Exact Recovery of Multichannel Sparse Blind Deconvolution via Gradient Descent
SIAM J. Imaging Sci. 2020 · Spotlight (top 3%) · Paper
Convolutional Phase Retrieval via Gradient Descent
IEEE Trans. Inf. Theory 2020 · Paper
A Geometric Analysis of Phase Retrieval
FoCM 2018 · Frontiers of Science Award in Mathematics (ICBS'24) · Paper
Complete Dictionary Recovery over the Sphere I: Overview and the Geometric Picture
IEEE Trans. Inf. Theory 2017 · Best Student Paper Award (SPARS 2015) · Paper
Finding a Sparse Vector in a Subspace: Linear Sparsity Using Alternating Directions
IEEE Trans. Inf. Theory 2016 · Paper
Abundance Estimation for Bilinear Mixture Models via Joint Sparse and Low-Rank Representation
IEEE TGRS 2014 · Paper · Code (.zip)