Publications by Research Topic
View the complete list in chronological order.
Papers are grouped by primary research topic, with each paper listed once. Within each subtopic, entries follow the order of the chronological list.
- Language Models, Watermarking and AI Security (19)
- Differential Privacy (39)
- Reinforcement Learning and Bandits (28)
- Online Learning and Distribution Shift (18)
- Dynamic Pricing and Resource Allocation (8)
- Deep Learning Theory and Efficient Optimization (12)
- Structured Estimation and Low-Dimensional Learning (15)
- Statistical Inference and Applications (7)
Language Models, Watermarking and AI Security
Efficient inference, alignment and model compression
- Gumbo: Gumbel Optimized High-Temperature Speculative Sampling
NeurIPS 2026 [available soon] - Learnable Chernoff Baselines for Inference-Time Alignment
NeurIPS 2026[pdf] - Not-a-Bandit: Provably No-Regret Drafter Selection in Speculative Decoding for LLMs
ICLR 2026 [arxiv] - ProxSparse: Regularized Learning of Semi-Structured Sparsity Masks for Pretrained LLMs
ICML 2025 [arxiv]
Watermarking, provenance and copyright protection
- Dataset Watermarking for Closed LLMs with Provable Detection
Manuscript [arxiv] - A Technical Report on “Erasing the Invisible”: The 2024 NeurIPS Competition on Stress Testing Image Watermarks
NeurIPS 2025 (Dataset and Benchmark Track). [available soon, Website] - Efficiently Identifying Watermarked Segments in Mixed-Source Texts
ACL 2025 [arxiv, code] - Permute-And-Flip: An Optimally Robust and Watermarkable Decoder for LLMs
ICLR 2025 [arxiv, code] - SoK: Watermarking for AI-Generated Content
IEEE S&P 2025. [arxiv] - Invisible Image Watermarks Are Provably Removable Using Generative AI
NeurIPS 2024 [arxiv, code] - Provable Robust Watermarking for AI-Generated Text
ICLR 2024 [arxiv, slides, code, demo] - CPR: Retrieval Augmented Generation for Copyright Protection
CVPR 2024. [arxiv] - Protecting Language Generation Models via Invisible Watermarking
ICML 2023. [arxiv] - Distillation-Resistant Watermarking for Model Protection in NLP
Findings of EMNLP 2022. [arxiv]
Reasoning benchmarks, implementation support and adversarial attacks
- DPrivBench: Benchmarking LLMs' Reasoning for Differential Privacy
NeurIPS 2026 (Benchmark and Dataset) [arxiv] - ConvexBench: Can LLMs Recognize Convex Functions?
ICML 2026 [arxiv] - DPCheatSheet: Using Worked and Erroneous LLM-usage Examples to Scaffold
Differential Privacy Implementation
Manuscript. [arxiv] - Weak-to-Strong Jailbreaking on Large Language Models
ICML 2025 [arxiv, code] -
Bullseye Polytope: A Scalable Clean-Label Poisoning Attack with Improved Transferability.
IEEE EuroS&P 2021. [arxiv]
Differential Privacy
Privacy foundations, accounting and individual guarantees
- Individual-Point Memorization:
Morris Scores, OA-KL Privacy, and Mutual Information
Manuscript [pdf] -
Purifying Approximate Differential Privacy with Randomized Post-processing
NeurIPS 2025. [arxiv] (*Spotlight) -
Privacy Profiles for Private Selection
ICML 2024. [arxiv] - Advancing Differential Privacy: Where We Are Now and Future Directions for Real-World Deployment
Harvard Data Science Review [paper, arxiv] - Generalized PTR: User-Friendly Recipes for Data-Adaptive Algorithms with Differential Privacy
AISTATS 2023. [arxiv] (*Plenary Oral Presentation) - Optimal Accounting of Differential Privacy via Characteristic Function
AISTATS 2022. [arxiv] - Privately Publishable Per-instance Privacy
NeurIPS 2021. [arxiv] - Improving Sparse Vector Technique with Renyi Differential Privacy
NeurIPS 2020. [paper,supplement ] - Poisson Subsampled Rényi Differential Privacy
ICML 2019. [paper,code] - Subsampled Rényi Differential Privacy and Analytical Moments Accountant
AISTATS 2019. [arxiv,code] (*Notable Paper Award, *Plenary oral presentation) - Improving Gaussian Mechanism for Differential Privacy: Analytical Calibration and Optimal Denoising
ICML 2018, Stockholm, Sweden. [arxiv] - Per-instance Differential Privacy
Journal of Confidentiality and Privacy. [pdf] - On-Average KL-Privacy and its equivalence to Generalization for Max-Entropy Mechanisms
Privacy in Statistical Databases. PSD'2016, Dubrovnik. [springer, arxiv] - Learning with Differential Privacy: Stability, Learnability and the Sufficiency and Necessity of ERM Principle
Journal of Machine Learning Research (JMLR), 2016. [arxiv]
Private learning, optimization and foundation models
- Private-RAG: Answering Multiple Queries with LLMs while Keeping Your Data Private
COLM 2026 [arxiv] -
Adapting to Linear Separable Subsets with Large-Margin in Differentially Private Learning
ICML 2025 [arxiv] - Neural Collapse meets Differential Privacy: Curious behaviors of NoisyGD with Near-Perfect Representation Learning
ICML 2024. [arxiv] (*Oral Presentation) - Differentially Private Bias-Term only Fine-tuning of Foundation Models
ICML 2024 [arxiv]. Short version appeared at NeurIPS'22 TSRML Workshop -
Tractable MCMC for Private Learning with Pure and Gaussian Differential Privacy
ICLR 2024. [arxiv] - Improved differentially private regression via gradient boosting
IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) [arxiv] -
Improving the Privacy and Practicality of Objective Perturbation for Differentially Private Linear Learners
NeurIPS 2023. [arxiv] - Automatic Clipping: Differentially Private Deep Learning Made Easier and Stronger
NeurIPS 2023. [arxiv]. Short version appeared at ICML'22 TPDP workshop.) - "Private Prediction Strikes back!" Private Kernelized Nearest Neighbors with Individual Renyi Filter
UAI 2023. [arxiv] (*Spotlight Presentation) - Differentially Private Optimization on Large Model at Small Cost
ICML 2023. [arxiv, code] - SeqPATE: Differentially Private Text Generation via Knowledge Distillation
NeurIPS 2022. [openreview] - Provably Confidential Language Modelling
NAACL 2022. [arxiv] (*Oral Presentation) - Mixed Differential Privacy in Computer Vision
CVPR 2022. [arxiv](*Oral Presentation) - Adaptive Private-K-Selection with Adaptive K and Application to Multi-label PATE
AISTATS 2022. [arxiv] - Voting-based Approaches For Differentially Private Federated Learning
Manuscript. [arxiv] -
Revisiting Model-Agnostic Private Learning: Faster Rates and Active Learning
Journal of Machine Learning Research. Shorter version appeared in AISTATS 2021. [arxiv] -
An end-to-end Differentially Private Latent Dirichlet Allocation Using a Spectral Algorithm
ICML 2020. [pdf, code] -
Private-kNN: Practical Differential Privacy for Computer Vision
CVPR 2020. [paper] - Revisiting differentially private linear regression: optimal and adaptive prediction and estimation in unbounded domain
UAI 2018, Monterey, CA[arxiv] (*Plenary oral presentation) - Privacy for Free: Posterior Sampling and Stochastic Gradient Monte Carlo
ICML 2015, Lille, France. [arxiv]
Private online learning, data valuation and data structures
- Improved Regret in Stochastic Decision-Theoretic Online Learning under Differential Privacy
International Conference on Algorithmic Learning Theory (ALT), 2026. [arxiv] - Threshold KNN-Shapley: A Linear-Time and Privacy-Friendly Approach to Data Valuation
NeurIPS 2023. [arxiv] (*Spotlight Presentation) - Differentially Private Linear Sketches: Efficient Implementations and Applications
NeurIPS 2022. [arxiv] - Differentially Private Subspace Clustering
NIPS 2015, Montreal, Canada. [pdf] - Fast Differential Private Matrix Factorization
RecSys'15, Vienna, Austria. [arxiv, code]
Reinforcement Learning and Bandits
Offline learning and off-policy evaluation
- On the Statistical Complexity for Offline and Low-Adaptive Reinforcement Learning with Structures
Accepted to Statistical Science [arxiv] - Offline Reinforcement Learning with Closed-Form Policy Improvement Operators
ICML 2023. [arxiv, code] - Offline Reinforcement Learning with Differentiable Function Approximation is Provably Efficient
ICLR 2023. [arxiv] - Offline Policy Evaluation for Reinforcement Learning with Adaptively Collected Data
Annual Learning for Dynamics & Control Conference (L4DC 2025) [arxiv] - Offline Stochastic Shortest Path: Learning, Evaluation and Towards Optimality
UAI 2022. [arxiv] - Near-optimal Offline Reinforcement Learning with Linear Representation: Leveraging Variance Information with Pessimism
ICLR 2022. [arxiv] - Towards Instance-Optimal Offline Reinforcement Learning with Pessimism
NeurIPS 2021. [arxiv] - Optimal Uniform OPE and Model-based Offline Reinforcement Learning in Time-Homogeneous, Reward-Free and Task-Agnostic Settings
NeurIPS 2021. [arxiv] - Near-Optimal Offline Reinforcement Learning via Double Variance Reduction
NeurIPS 2021. [arxiv] -
Near Optimal Provable Uniform Convergence in Offlin Policy Evaluation for Reinforcement Learning
AISTATS 2021. (*Plenary oral presentation) [arxiv] -
Asymptotically Efficient Off-Policy Evaluation for Tabular Reinforcement Learning
AISTATS 2020. [arxiv] - Towards Optimal Off-Policy Evaluation for Reinforcement Learning with Marginalized Importance Sampling
NeurIPS 2019. [arxiv] - Imitation Regularized Offline Learning
AISTATS 2019. [arxiv] - Optimal and Adaptive Off-Policy Evaluation in Contextual Bandits
ICML 2017. [paper supp slides]
Exploration, limited adaptivity, nonstationarity and games
- NetworkGym: Reinforcement Learning Environments for Multi-Access Traffic Management in Network Simulation
NeurIPS 2024 (Dataset and Benchmark Track). [arxiv, github] - Near-Optimal Reinforcement Learning with Self-Play under Adaptivity Constraints
ICML 2024. [arxiv] - Improving Sample Efficiency of Model-Free Algorithms for Zero-Sum Markov Games
ICML 2024. [arxiv] - Communication-Efficient Federated Non-Linear Bandit Optimization
ICLR 2024 [arxiv] - Towards General Function Approximation in Nonstationary Reinforcement Learning
IEEE Journal on Selected Areas in Information Theory. [ieee] - Posterior Sampling with Delayed Feedback for Reinforcement Learning with Linear Function Approximation
NeurIPS 2023. [arxiv] - No-Regret Linear Bandits beyond Realizability
UAI 2023. [arxiv] - Non-stationary Reinforcement Learning under General Function Approximation
ICML 2023. [arxiv] - Near-Optimal Deployment Efficiency in Reward-Free Reinforcement Learning with Linear Function Approximation
ICLR 2023. [arxiv] - Sample-Efficient Reinforcement Learning with loglog(T) Switching Cost
ICML 2022. [arxiv] - Provably Efficient Q-Learning with Low Switching Cost
NeurIPS 2019. [arxiv]
Differentially private reinforcement learning
- Differentially Private Reinforcement Learning with Self-Play
NeurIPS 2024. [arxiv] - Offline Reinforcement Learning with Differential Privacy
NeurIPS 2023. [arxiv] - Near-Optimal Differentially Private Reinforcement Learning
AISTATS 2023 [arxiv]
Online Learning and Distribution Shift
Adaptive regret, nonstationary optimization and control
- Universal Dynamic Portfolio
Manuscript [arxiv] - A second-order regret bound for NormalHedge
NeurIPS 2026 [arxiv] - Non-stationary Online Learning with Memory and Non-stochastic Control
Journal of Machine Learning Research (JMLR), 2023 [pdf] Shorter version appeared in AISTATS 2022. - Second Order Path Variationals in Non-Stationary Online Learning
AISTATS 2023. [arxiv] - Optimal Dynamic Regret in LQR Control
NeurIPS 2022. [arxiv] - Optimal Dynamic Regret in Proper Online Learning with Strongly Convex Losses and Beyond
AISTATS 2022. [arxiv] - Optimal Dynamic Regret in Exp-Concave Online Learning
COLT 2021. [arxiv] (*Best Student Paper Award) - Non-stationary Stochastic Optimization under Lp,q-Variation Measures
Operations Research. [informs_online, arxiv]
Distribution shift and label shift adaptation
- Adaptive Estimation and Learning under Temporal Distribution Shift
ICML 2025 [arxiv] - Adapting to Online Distribution Shifts in Deep Learning: A Black-Box Approach
AISTATS 2025 [arxiv] - Online Feature Updates Improve Online (Generalized) Label Shift Adaptation
NeurIPS 2024. [arxiv] - Online Label Shift: Optimal Dynamic Regret meets Practical Algorithms
NeurIPS 2023. [arxiv, code] (*Spotlight Presentation) -
Domain Adaptation with Conditional Distribution Matching and Generalized Label Shift
NeurIPS 2020. [arxiv] - Detecting and Correcting for Label Shift with Black Box Predictors
ICML 2018, Stockholm, Sweden. [arxiv]
Online nonparametric estimation and forecasting
- AKORN: Adaptive Knots generated Online for RegressioN Splines
ICML 2025[pdf] - An Optimal Reduction of TV-Denoising to Adaptive Online Learning
AISTATS 2021. [arxiv] - Adaptive Online Estimation of Piecewise Polynomial Trends
NeurIPS 2020. [arxiv] - Online Forecasting of Total Variation-Bounded Sequences
NeurIPS 2019. [arxiv]
A short version appeared at ICML'19 Time Series Workshop. (*Best Paper Honorable Mention)
Dynamic Pricing and Resource Allocation
- Optimal Contextual Pricing under Agnostic Non-Lipschitz Demand
NeurIPS 2026 [arxiv] - Dynamic Pricing with Adversarially-Censored Demands.
Conference on Web and Internet Economics (WINE 2025)[arxiv] - Joint Pricing and Resource Allocation: An Optimal Online-Learning Approach
Manuscript [arxiv] - Pricing with Contextual Elasticity and Heteroscedastic Valuation.
ICML 2024[arxiv] (*Spotlight) - Doubly Fair Dynamic Pricing
AISTATS 2023 [arxiv] - Non-stationary Contextual Pricing with Safety Constraints
Transaction of Machine Learning Research [openreview] - Towards Agnostic Feature-based Dynamic Pricing: Linear Policies vs Linear Valuation with Unknown Noise
AISTATS 2022. [arxiv] (*Plenary Oral Presentation) - Logarithmic Regret in Feature-Based Dynamic Pricing
NeurIPS 2021. [arxiv] (*Spotlight Presentation)
Deep Learning Theory and Efficient Optimization
Generalization, stability and nonparametric perspectives
- Does Sparse Connectivity Improve Generalization? Convolutional Networks Below the Edge of Stability
NeurIPS 2026 [arxiv] - Generalization Below the Edge of Stability: The Role of Data Geometry
ICLR 2026 [arxiv] - Stable Minima of ReLU Neural Networks Suffer from the Curse of Dimensionality: The Neural Shattering Phenomenon
NeurIPS 2025. [arxiv] (*Spotlight) - Stable Minima Cannot Overfit in Univariate ReLU Networks: Generalization by Large Step Sizes
NeurIPS 2024. [arxiv] (*Spotlight) - Nonparametric Classification on Low Dimensional Manifolds using Overparameterized Convolutional Residual Networks
NeurIPS 2024 [arxiv] - Deep Learning meets Nonparametric Regression: Are Weight-Decayed DNNs Locally Adaptive?
ICLR 2023. [arxiv]
Sparsity, quantization, global and distributed optimization
- A Proximal Operator for Inducing 2:4-Sparsity
Transactions on Machine Learning Research [arxiv] - Global Optimization with Parametric Function Approximation
ICML 2023. [arxiv] - ProxQuant: Quantized Neural Networks via Proximal Operators.
ICLR 2019. [arxiv] - signSGD: compressed optimisation for non-convex problems
ICML 2018, Stockholm, Sweden. [arxiv] - Parallel and Distributed Block-Coordinate Frank-Wolfe Algorithms
ICML 2016, New York City. [pdf][supplement] - DiFacto --- Distributed Factorization Machines (*Best Paper Honorable Mention)
WSDM 2016, San Francisco. [pdf]
Structured Estimation and Low-Dimensional Learning
Trend filtering, total variation and nonparametric statistics
- Multivariate Trend Filtering for Lattice Data
Annals of Statistics [arxiv, DOI] - A Higher-Order Kolmogorov-Smirnov Test
AISTATS 2019. [arxiv] (*Plenary oral presentation) - Higher-Order Total Variation Classes on Grids:
Minimax Theory and Trend Filtering Methods
NIPS 2017. [paper,supp，poster] - Attributing Hacks with Survival Trend Filtering
Electronic Journal of Statistics[ paper slides]
"Attributing Hacks" in AISTATS'17 (*Plenary oral presentation) - Total Variation Classes Beyond 1d: Minimax Rates, and the
Limitations of Linear Smoothers
NIPS 2016, Barcelona. [arxiv] - Graph Sparsification Approaches for Large-Scale Laplacian Smoothing
AISTATS 2016, Cadiz, Spain. [pdf, supplementary] - Trend Filtering on Graphs
Journal of Machine Learning Research (JMLR), 2016. [jmlr]
AISTATS 2015, San Diego. [conf] - The Falling Factorial Basis and Its Statistical Applications
ICML 2014, Beijing. [paper, Supplementary, code]
Subspace clustering, matrix completion and robust recovery
- Graph Connectivity in Noisy Sparse Subspace Clustering
AISTATS 2016, Cadiz, Spain. [arxiv] - A Theoretical Analysis of Noisy Sparse Subspace Clustering for Dimensionality-reduced Data
IEEE Transaction on Information Theory[ieee,arxiv]
ICML 2015, Lille, France. [conf] - Provable Subspace Clustering: When LRR meets SSC
IEEE Transaction on Information Theory.[fullpaper,arxiv]
NIPS 2013. Selected for spotlight presentation. [conf, demo code] - Noisy Sparse Subspace Clustering
Journal of Machine Learning Research (2016). [pdf]
ICML 2013, Atlanta [paper, Supplementary, code, Video] - Practical Matrix Completion and Corruption Recovery using Proximal Alternating Robust Subspace Minimization
International Journal of Computer Vision (IJCV), 2014. [Springerlink,DemoCode] - Block-Sparse RPCA for Salient Motion Detection
Pattern Analysis and Machine Intelligence (PAMI), 2014. [IEEEXplore] - Stability of Matrix Factorization for Collaborative Filtering
ICML 2012, Edinburgh. [paper, Supplementary, Slides , Video]
Statistical Inference and Applications
Adaptive data analysis, crowdsourcing and ecological inference
- Doubly Robust Crowdsourcing
Journal of Artificial Intelligence Research [pdf] - Understanding the 2016 US Presidential Election using ecological inference and distribution regression with census microdata
In preparation. [arxiv] - A Minimax Theory for Adaptive Data Analysis
In preparation. [arxiv] - Who supported Obama in 2012? Ecological inference through distribution regression (*Best Student Paper)
KDD 2015, Sydney, Australia. [pdf]
Time series forecasting and vision
- Inter-Series Attention Model for COVID-19 Forecasting
SDM 2021. [arxiv] - Enhancing the Locality and Breaking the Memory Bottleneck of Transformer on Time Series Forecasting
NeurIPS 2019. [arxiv] - An efficient algorithm for UAV indoor pose
estimation using vanishing geometry
MVA 2011, Nara, Japan [paper, Poster]