Qi Lei
Assistant Professor of Mathematics and Data Science
Studies machine learning theory, including representation learning, distribution shift and data privacy in deep learning.
Qi Lei's lab at NYU works on learning theory & optimization, trustworthy & safe ML and data management & big data. 32 papers since 2023, 18 at top venues, mostly NeurIPS and AISTATS. 5 PhD students. Applications close .
At a glance
- papers, 2023–2026
- 8/ year
- at T1 venues
- 56%
- PhD students
- 5
- MS students
- 0
- alumni listed
- —
Papers by yearfrom the lab homepage, 2023–2026
- 5 papers2026 · T1 1 · collecting
- 12 papers2025 · T1 7
- 5 papers2024 · T1 4
- 10 papers2023 · T1 6
T1 venuesOtherVenues: NeurIPS, AISTATS, ICML, Findings of EMNLP, ICLR
Research focuspaper titles that use each topic's words
- Learning theory & optimization12 papers
- Data management & big data5 papers
- Trustworthy & safe ML0 papers
- data privacy0 papers
- shift0 papers
Paper list
From the lab homepage.
20265 papers
- to appear at ICMLA Task-Centric Theory for Iterative Self-Improvement with Easy-to-Hard Curricula (opens in a new tab)
- —Data Reconstruction: Identifiability and Optimization with Sample Splitting, to appear at ICML 2026
- ICLR (T1)Does Weak-to-strong Generalization Happen under Spurious Correlations? (opens in a new tab)
- CPALSuperclass-Guided Representation Disentanglement for Spurious Correlation Mitigation (opens in a new tab)
- LIT Workshop@ICLRWhen does Chain-of-Thought Help: A Markovian Perspective (opens in a new tab)
202512 papers
- ICLR (T1)Beyond Interpretability: The Gains of Feature Monosemanticity on Model Robustness (opens in a new tab)
- UAI (T1)Beyond Invisibility: Learning Robust Visible Watermarks for Stronger Copyright Protection (opens in a new tab)
- ICCV highlightBeyond Losses Reweighting: Empowering Multi-Task Learning via the Generalization Perspective (opens in a new tab)
- CPALBridging Domain Adaptation and Graph Neural Networks: A Tensor-Based Framework for Effective Label Propagation (opens in a new tab)
- AISTATS (T1)Bridging Domains with Approximately Shared Features (opens in a new tab)
- —Data Reconstruction Attacks and Defenses: A Systematic Evaluation, AISTATS 2025
- ICML (T1)Discrepancies are Virtue: Weak-to-Strong Generalization through Lens of Intrinsic Dimension (opens in a new tab)
- AISTATS (T1)Elastic Representation: Mitigating Spurious Correlations for Group Robustness (opens in a new tab)
- CPALGreedy Output Approximation: Towards Efficient Structured Pruning for LLMs Without Retraining (opens in a new tab)
- NeurIPS (T1)Hyperparameter Transfer Enables Consistent Gains of Matrix-Preconditioned Optimizers Across Scales (opens in a new tab)
- NeurIPS (T1)Performative Risk Control: Calibrating Models for Reliable Deployment under Performativity (opens in a new tab)
- Findings of EMNLPThink Twice, Generate Once: Enhancing LLMs Safety via Progressive Self-Reflection (opens in a new tab)
20245 papers
- ICML (T1)‘‘An Information-Theoretic Analysis of In-Context Learning" (opens in a new tab)
- ICML (T1)‘‘Controllable Prompt Tuning For Balancing Group Distributional Robustness" (opens in a new tab)
- CPAL‘‘Exploring Minimally Sufficient Representation in Active Learning through Label-Irrelevant Patch Augmentation" (opens in a new tab)
- NeurIPS (T1)Sketchy Moment Matching: Toward Fast and Provable Data Selection for Finetuning (opens in a new tab)
- NeurIPS (T1)Stochastic Zeroth-Order Optimization under Strongly Convexity and Lipschitz Hessian: Minimax Sample Complexity (opens in a new tab)
202310 papers
- FL@FM-NeurIPS‘‘Beyond Gradient and Priors in Privacy Attacks: Leveraging Pooler Layer Inputs of Language Models in Federated Learning" (opens in a new tab)
- Findings of EMNLP‘‘Breaking through Deterministic Barriers: Randomized Pruning Mask Generation and Selection" (opens in a new tab)
- NeurIPS (T1)‘‘Cluster-aware Semi-supervised Learning: Relational Knowledge Distillation Provably Learns Clustering" (opens in a new tab)
- —‘‘Optimal Sample Complexity Bounds for Non-convex Optimization under Kurdyka-Lojasiewicz Condition", AISTATS 2023
- ICML (T1)‘‘Optimization for Amortized Inverse Problems" (opens in a new tab)
- AISTATS (T1)‘‘Provable Hierarchy-Based Meta-Reinforcement Learning" (opens in a new tab)
- AISTATS (T1)‘‘Reconstructing Training Data from Model Gradient, Provably" (opens in a new tab)
- NeurIPS (T1)‘‘Sample Complexity for Quadratic Bandits: Hessian Dependent Bounds and Optimal Algorithms" (opens in a new tab)
- AISTATS (T1)‘‘Sample Efficiency of Data Augmentation Consistency Regularization" (opens in a new tab)
- EMNLP-Main‘‘Towards Robust Pruning: An Adaptive Knowledge-Retention Pruning Strategy for Language Models" (opens in a new tab)
Where alumni went
Alumni are not listed on the lab homepage.
Apply to NYUFall 2027
Computer Science PhD (Courant / Tandon)
- Round
- Single round (December)
- Opens
- —
- GRE
- See program page
- English
- See program page
- Fee
- See program page
Not yet confirmed for this intake.
Similar labsnearest by research summary
- CHETH Zurich
Niao He
- Learning theory & optimization
- CNZhejiang
Haobo Wang
78 T1 - USMIT
Tomaso Poggio
- Computational neuroscience & cognition
- CNTsinghua
Jian Li
- Learning theory & optimization
- CNTsinghua
Yu Liu
- Reinforcement learning
- CAToronto
Ohad Shamir
- Learning theory & optimization
Your notes
Save this lab to keep notes.
Data
Some numbers could not be verified:
- The alumni list could not be found on the lab site.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Not collected
Tiers are venue classifications, not lab rankings.
NYUcloses