Research

Research Interests

  • High-dimensional statistics and learning
  • Optimization and deep learning theory

Publications (*: equal contribution)

  1. Gradient Descent on Two ReLU Neurons: Global Landscape and Bifurcation Dynamics
    Binghua Li*, Mengzhe Li*, Denny Wu, and Tianhao Wang
    In Advances in Neural Information Processing Systems (NeurIPS), 2026
    Presented at ICML 2026 Workshop on High-dimensional Learning Dynamics (HiLD)
    arXiv
  2. Towards Understanding Momentum Acceleration in River-Valley Loss Landscape
    Miao Lu, Zeyu Bian, Kaiyue Wen, Beining Wu, Siyu Chen,  Tianhao Wang and 1 more author
    In Advances in Neural Information Processing Systems (NeurIPS), 2026
    arXiv
  3. INFUSER: Influence-Guided Self-Evolution Improves Reasoning
    Siyu Chen, Miao Lu, Beining Wu, Heejune Sheen, Fengzhuo Zhang,  Shuangning Li and 4 more authors
    In Advances in Neural Information Processing Systems (NeurIPS), 2026
    arXiv
  4. On Universality of Non-Separable Approximate Message Passing Algorithms
    Max Lovig, Tianhao Wang, and Zhou Fan
    Annals of Statistics (to appear), 2026
    arXiv
  5. Your Reasoning Benchmark May Not Test Reasoning: Revealing Perception Bottleneck in Abstract Reasoning Benchmarks
    Xinhe Wang, Jin Huang, Xingjian Zhang, Tianhao Wang, and Jiaqi W. Ma
    In Annual Meeting of the Association for Computational Linguistics (ACL), 2026
    arXiv Link
  6. Through the Judge’s Eyes: Inferred Thinking Traces Improve Reliability of LLM Raters
    Xingjian Zhang, Tianhong Gao, Suliang Jin, Tianhao Wang, Teng Ye,  Eytan Adar and 1 more author
    Transactions on Machine Learning Research, 2026
    arXiv
  7. A Tale of Two Geometries: Adaptive Optimizers and Non-Euclidean Descent
    Shuo Xie, Tianhao Wang, Beining Wu, and Zhiyuan Li
    In International Conference on Learning Representations (ICLR), 2026
    arXiv
  8. Taming Polysemanticity in LLMs: Provable Feature Recovery via Sparse Autoencoders
    Siyu Chen, Heejune Sheen, Xuyuan Xiong, Tianhao Wang, and Zhuoran Yang
    In International Conference on Learning Representations (ICLR), 2026
    arXiv
  9. Structured Preconditioners in Adaptive Optimization: A Unified Analysis
    Shuo Xie, Tianhao Wang, Sashank Reddi, Sanjiv Kumar, and Zhiyuan Li
    In International Conference on Machine Learning (ICML), 2025
    arXiv
  10. Can Neural Networks Achieve Optimal Computational-statistical Tradeoff? An Analysis on Single-Index Model
    Siyu Chen*, Beining Wu*, Miao Lu, Zhuoran Yang, and Tianhao Wang
    In International Conference on Learning Representations (ICLR), 2025  (Oral)
    Presented at NeurIPS 2024 Workshop on Mathematics of Modern Machine Learning
    arXiv Link
  11. How well can Transformers emulate in-context Newton’s method?
    Angeliki Giannou, Liu Yang, Tianhao Wang, Dimitris Papailiopoulos, and Jason D. Lee
    In International Conference on Artificial Intelligence and Statistics (AISTATS), 2025
    Presented at ICLR 2024 Workshop on Bridging the Gap Between Practice and Theory in Deep Learning
    arXiv
  12. Unveiling Induction Heads: Provable Training Dynamics and Feature Learning in Transformers
    Siyu Chen, Heejune Sheen, Tianhao Wang, and Zhuoran Yang
    In Advances in Neural Information Processing Systems (NeurIPS), 2024
    Presented at ICML 2024 Workshop on Theoretical Foundations of Foundation Models
    arXiv Link
  13. Approximate Message Passing for orthogonally invariant ensembles: Multivariate non-linearities and spectral initialization
    Xinyi Zhong*, Tianhao Wang*, and Zhou Fan
    Information and Inference: A Journal of the IMA, 2024
    arXiv Link
  14. Universality of Approximate Message Passing algorithms and tensor networks
    Tianhao Wang, Xinyi Zhong, and Zhou Fan
    Annals of Applied Probability, 2024
    arXiv Link
  15. Training dynamics of multi-head softmax attention for in-context learning: emergence, convergence, and optimality
    Siyu Chen, Heejune Sheen, Tianhao Wang, and Zhuoran Yang
    Conference on Learning Theory (COLT), 2024
    Presented at ICLR 2024 Workshop on Bridging the Gap Between Practice and Theory in Deep Learning
    arXiv
  16. Maximum likelihood for high-noise group orbit estimation and single-particle cryo-EM
    Zhou Fan, Roy R. Lederman, Yi Sun, Tianhao Wang, and Sheng Xu
    Annals of Statistics, 2024
    arXiv Link
  17. The Marginal Value of Momentum for Small Learning Rate SGD
    Runzhe Wang, Sadhika Malladi, Tianhao Wang, Kaifeng Lyu, and Zhiyuan Li
    In International Conference on Learning Representations (ICLR), 2024
    arXiv Link
  18. Noise-adaptive Thompson sampling for linear contextual bandits
    Ruitu Xu, Yifei Min, and Tianhao Wang
    In Advances in Neural Information Processing Systems (NeurIPS), 2023
    Link
  19. Cooperative multi-Agent reinforcement learning: asynchronous communication and linear function approximation
    Yifei Min, Jiafan He, Tianhao Wang, and Quanquan Gu
    In International Conference on Machine Learning (ICML), 2023
    arXiv Link
  20. Finding regularized competitive equilibria of heterogeneous agent macroeconomic models via reinforcement learning
    Ruitu Xu, Yifei Min, Tianhao Wang, Michael I. Jordan, Zhaoran Wang, and Zhuoran Yang
    In International Conference on Artificial Intelligence and Statistics (AISTATS), 2022
    Link
  21. Fast mixing of stochastic gradient descent with normalization and weight decay
    Zhiyuan Li, Tianhao Wang, and Dingli Yu
    In Advances in Neural Information Processing Systems (NeurIPS), 2022
    Link
  22. Learn to match with no regret: Reinforcement learning in Markov matching markets
    Yifei Min, Tianhao Wang, Ruitu Xu, Zhaoran Wang, Michael I Jordan, and Zhuoran Yang
    In Advances in Neural Information Processing Systems (NeurIPS), 2022  (Oral)
    arXiv Link
  23. A simple and provably efficient algorithm for asynchronous federated contextual linear bandits
    Jiafan He*, Tianhao Wang*, Yifei Min*, and Quanquan Gu
    In Advances in Neural Information Processing Systems (NeurIPS), 2022
    arXiv Link
  24. Implicit bias of gradient descent on reparametrized models: On equivalence to mirror descent
    Zhiyuan Li*, Tianhao Wang*, Jason D. Lee, and Sanjeev Arora
    In Advances in Neural Information Processing Systems (NeurIPS), 2022
    Abridged version accepted for a contributed talk to ICML 2022 Workshop on Continuous time methods for machine learning
    arXiv Poster Slides
  25. Learning stochastic shortest path with linear function approximation
    Yifei Min, Jiafan He, Tianhao Wang, and Quanquan Gu
    In International Conference on Machine Learning (ICML), 2022
    arXiv Link Poster Slides
  26. What happens after SGD reaches zero loss?–A mathematical framework
    Zhiyuan Li, Tianhao Wang, and Sanjeev Arora
    In International Conference on Learning Representations (ICLR), 2022  (Spotlight)
    arXiv Link Poster Slides
  27. North American biliary stricture management strategies in children after liver transplantation: a multicenter analysis from the society of pediatric liver transplantation (SPLIT) registry
    Pamela L Valentino, Tianhao Wang, Veronika Shabanova, Vicky Lee Ng, John C Bucuvalas,  Amy G Feldman and 5 more authors
    Liver Transplantation, 2022
    Link
  28. Variance-aware off-policy evaluation with linear function approximation
    Yifei Min*, Tianhao Wang*, Dongruo Zhou, and Quanquan Gu
    In Advances in neural information processing systems (NeurIPS), 2021
    arXiv Link Poster Slides
  29. Provably efficient reinforcement learning with linear function approximation under adaptivity constraints
    Tianhao Wang*, Dongruo Zhou*, and Quanquan Gu
    In Advances in Neural Information Processing Systems (NeurIPS), 2021
    arXiv Link Poster Slides
  30. Likelihood landscape and maximum likelihood estimation for the discrete orbit recovery model
    Zhou Fan, Yi Sun, Tianhao Wang, and Yihong Wu
    Communications on Pure and Applied Mathematics, 2022
    arXiv Link
  31. Continuous and discrete-time accelerated stochastic mirror descent for strongly convex functions
    Pan Xu*, Tianhao Wang*, and Quanquan Gu
    In International Conference on Machine Learning (ICML), 2018
    Link
  32. Accelerated stochastic mirror descent: From continuous-time dynamics to discrete-time algorithms
    Pan Xu*, Tianhao Wang*, and Quanquan Gu
    In International Conference on Artificial Intelligence and Statistics (AISTATS), 2018
    Link

Preprints (*: equal contribution)

  1. LLM Agents Enable User-Governed Personalization Beyond Platform Boundaries
    Jiacheng Lin, Kun Qian, Arvind Srinivasan, Tian Wang, Fang Han,  Changran Hu and 12 more authors
    arXiv:2605.09794, 2026
    arXiv
  2. Provable Benefit of SignGD: A Minimal Model Under Heavy-Tailed Class Imbalance
    Robin Yadav, Shuo Xie, Tianhao Wang, and Zhiyuan Li
    arXiv:2512.00763, 2025
    Oral presentation at NeurIPS 2025 Workshop on Optimization for Machine Learning (OPT 2025)
    arXiv Link
  3. Honesty over Accuracy: Trustworthy Language Models through Reinforced Hesitation
    Mohamad Amin Mohamadi, Tianhao Wang, and Zhiyuan Li
    arXiv:2511.11500, 2025
    arXiv
  4. Implicit regularization of gradient flow on one-layer softmax attention
    Heejune Sheen, Siyu Chen, Tianhao Wang, and Harrison H. Zhou
    arXiv:2403.08699, 2024
    Presented at ICLR 2024 Workshop on Bridging the Gap Between Practice and Theory in Deep Learning
    arXiv