Yu-Xiang Wang
Associate Professor
Designs efficient machine learning algorithms with provable guarantees, including differential privacy, reinforcement learning and watermarking of generative AI.
Yu-Xiang Wang's lab at UCSD works on reinforcement learning, generative & foundation models and efficient training & inference. 67 papers since 2023, 46 at top venues, mostly NeurIPS and ICML. 6 PhD students. 17 alumni, most often to Amazon and Berkeley. Xuandong worked on responsible use of generative AI. Applications close .
At a glance
- papers, 2023–2026
- 17/ year
- at T1 venues
- 69%
- PhD students
- 6
- MS students
- 0
- alumni listed
- 17
Papers by yearfrom the lab homepage, 2023–2026
- 11 papers2026 · T1 9 · collecting
- 16 papers2025 · T1 9
- 19 papers2024 · T1 12
- 21 papers2023 · T1 16
T1 venuesOtherVenues: NeurIPS, ICML, ICLR, AISTATS, UAI
Research focuspaper titles that use each topic's words
- Reinforcement learning24 papers
- Generative & foundation models4 papers
- Efficient training & inference3 papers
- watermarking3 papers
- provable guarantees0 papers
Paper list
From the lab homepage.
202611 papers
- NeurIPS (T1)A second-order regret bound for NormalHedge (opens in a new tab)
- ICML (T1)ConvexBench: Can LLMs Recognize Convex Functions? (opens in a new tab)
- NeurIPS (T1)Does Sparse Connectivity Improve Generalization? Convolutional Networks Below the Edge of Stability (opens in a new tab)
- NeurIPS (T1)DPrivBench: Benchmarking LLMs' Reasoning for Differential Privacy (opens in a new tab)
- ICLR (T1)Generalization Below the Edge of Stability: The Role of Data Geometry (opens in a new tab)
- NeurIPS (T1)Gumbo: Gumbel Optimized High-Temperature Speculative Sampling
- International Conference on Algorithmic Learning Theory (ALT)Improved Regret in Stochastic Decision-Theoretic Online Learning under Differential Privacy (opens in a new tab)
- NeurIPS (T1)Learnable Chernoff Baselines for Inference-Time Alignment (opens in a new tab)
- ICLR (T1)Not-a-Bandit: Provably No-Regret Drafter Selection in Speculative Decoding for LLMs (opens in a new tab)
- NeurIPS (T1)Optimal Contextual Pricing under Agnostic Non-Lipschitz Demand (opens in a new tab)
- COLMPrivate-RAG: Answering Multiple Queries with LLMs while Keeping Your Data Private (opens in a new tab)
202516 papers
- TMLR (T1)A Proximal Operator for Inducing 2:4-Sparsity (opens in a new tab)
- NeurIPS (Dataset and Benchmark Track). [, ]A Technical Report on “Erasing the Invisible”: The 2024 NeurIPS Competition on Stress Testing Image Watermarks
- ICML (T1)Adapting to Linear Separable Subsets with Large-Margin in Differentially Private Learning (opens in a new tab)
- AISTATS (T1)Adapting to Online Distribution Shifts in Deep Learning: A Black-Box Approach (opens in a new tab)
- ICML (T1)Adaptive Estimation and Learning under Temporal Distribution Shift (opens in a new tab)
- ICML (T1)AKORN: Adaptive Knots generated Online for RegressioN Splines (opens in a new tab)
- WINEDynamic Pricing with Adversarially-Censored Demands. (opens in a new tab)
- ACL [, ]Efficiently Identifying Watermarked Segments in Mixed-Source Texts (opens in a new tab)
- L4DCOffline Policy Evaluation for Reinforcement Learning with Adaptively Collected Data (opens in a new tab)
- Statistical ScienceOn the Statistical Complexity for Offline and Low-Adaptive Reinforcement Learning with Structures (opens in a new tab)
- ICLR [, ]Permute-And-Flip: An Optimally Robust and Watermarkable Decoder for LLMs (opens in a new tab)
- ICML (T1)ProxSparse: Regularized Learning of Semi-Structured Sparsity Masks for Pretrained LLMs (opens in a new tab)
202419 papers
- Harvard Data Science Review [, ]Advancing Differential Privacy: Where We Are Now and Future Directions for Real-World Deployment (opens in a new tab)
- ICLR (T1)Communication-Efficient Federated Non-Linear Bandit Optimization (opens in a new tab)
- CVPR (T1)CPR: Retrieval Augmented Generation for Copyright Protection (opens in a new tab)
- ICML . Short version appeared atDifferentially Private Bias-Term only Fine-tuning of Foundation Models (opens in a new tab)
- NeurIPS (T1)Differentially Private Reinforcement Learning with Self-Play (opens in a new tab)
- SaTMLImproved differentially private regression via gradient boosting (opens in a new tab)
- ICML (T1)Improving Sample Efficiency of Model-Free Algorithms for Zero-Sum Markov Games (opens in a new tab)
- NeurIPS [, ]Invisible Image Watermarks Are Provably Removable Using Generative AI (opens in a new tab)
- ICML (T1)Near-Optimal Reinforcement Learning with Self-Play under Adaptivity Constraints (opens in a new tab)
- NeurIPS (Dataset and Benchmark Track). [, ]NetworkGym: Reinforcement Learning Environments for Multi-Access Traffic Management in Network Simulation (opens in a new tab)
- ICML (T1)Neural Collapse meets Differential Privacy: Curious behaviors of NoisyGD with Near-Perfect Representation Learning (opens in a new tab)
- NeurIPS (T1)Nonparametric Classification on Low Dimensional Manifolds using Overparameterized Convolutional Residual Networks (opens in a new tab)
202321 papers
- UAI (T1)"Private Prediction Strikes back!" Private Kernelized Nearest Neighbors with Individual Renyi Filter (opens in a new tab)
- NeurIPS . . Short version appeared at ICML'22 TPDP workshop.)Automatic Clipping: Differentially Private Deep Learning Made Easier and Stronger (opens in a new tab)
- ICLR (T1)Deep Learning meets Nonparametric Regression: Are Weight-Decayed DNNs Locally Adaptive? (opens in a new tab)
- ICML . [, ]Differentially Private Optimization on Large Model at Small Cost (opens in a new tab)
- AISTATS (T1)Doubly Fair Dynamic Pricing (opens in a new tab)
- AISTATS (T1)Generalized PTR: User-Friendly Recipes for Data-Adaptive Algorithms with Differential Privacy (opens in a new tab)
- ICML (T1)Global Optimization with Parametric Function Approximation (opens in a new tab)
- NeurIPS (T1)Improving the Privacy and Practicality of Objective Perturbation for Differentially Private Linear Learners (opens in a new tab)
- ICLR (T1)Near-Optimal Deployment Efficiency in Reward-Free Reinforcement Learning with Linear Function Approximation (opens in a new tab)
- AISTATS (T1)Near-Optimal Differentially Private Reinforcement Learning (opens in a new tab)
- UAI (T1)No-Regret Linear Bandits beyond Realizability (opens in a new tab)
- Journal of Machine Learning Research (JMLR), Shorter version appeared in AISTATSNon-stationary Online Learning with Memory and Non-stochastic Control (opens in a new tab)
Where alumni went17 listed, 57% to industry
Industry 8Academia 6
Most often to
- Amazon
- Berkeley. Xuandong worked on responsible use of generative AI
- ByteDance
Also to
- Carnegie Mellon University
- Nanjing University
- Northeastern University. Rachel worked on differential privacy and differentially private machine learning
Apply to UCSDFall 2027
UCSD CSE Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- See program page
- English
- See program page
- Fee
- See program page
Not yet confirmed for this intake.
Similar labsnearest by research summary
- DETUM
Reinhard Heckel
- Signal processing & information theory
- USMIT
Tomaso Poggio
- Computational neuroscience & cognition
- USGeorgia Tech
Kai Wang
- Reasoning & neuro-symbolic AI
- CHEPFL
Pierre Vandergheynst
- Signal processing & information theory
- CNSJTU
Dequan Wang
13 T1 - SGNTU
Lihui Chen
- Recognition, detection & segmentation
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
UCSDcloses