publications

Preprints and publications by Jianhao Ma.

2026

  1. A lower bound for stepsize-based acceleration of gradient descent
    Jianhao Ma and Yuxin Chen
    arXiv preprint, 2026
    arXiv HTML
  2. WSqD: A Horizon-Free Learning Rate Schedule for Large Model Training
    Jianhao Ma and Yuxin Chen
    arXiv preprint, 2026
    arXiv HTML
  3. Towards Understanding the Power and Limits of the Muon Optimizer: A River-Valley Perspective
    Tianqi Shen, Jinji Yang, Runze Shi, Jianhao Ma, Jiaye Teng, and Ziye Ma
    arXiv preprint, 2026
    arXiv HTML
  4. Understanding Quantization-Aware Training: Gradients at Quantized Weights Bias to the Low-Loss Basin
    Hanyang Li, Jianhao Ma, and Ying Cui
    arXiv preprint, 2026
    arXiv HTML
  5. Preconditioning Benefits of Spectral Orthogonalization in Muon
    Jianhao Ma, Yu Huang, Yuejie Chi, and Yuxin Chen
    arXiv preprint, 2026
    arXiv HTML
  6. Certifying Optimality in Nonconvex Robust PCA
    Pinxi Gong, Lexiao Lai, and Jianhao Ma
    arXiv preprint, 2026
    arXiv HTML
  7. LIBRA: Language Model Informed Bandit Recourse Algorithm for Personalized Treatment Planning
    Junyu Cao, Ruijiang Gao, Esmaeil Keyvanshokooh, and Jianhao Ma
    arXiv preprint, 2026
    arXiv HTML
  8. Sparse Mean Estimation in Adversarial Settings via Incremental Learning
    Jianhao Ma, Rui Ray Chen, Yinghui He, Salar Fattahi, and Wei Hu
    Transactions on Machine Learning Research, 2026
    arXiv HTML

2025

  1. Quantization through Piecewise-Affine Regularization: Optimization and Statistical Guarantees
    Jianhao Ma and Lin Xiao
    arXiv preprint, 2025
    arXiv HTML
  2. Implicit Regularization of Infinitesimally-perturbed Gradient Descent Toward Low-dimensional Solutions
    Jianhao Ma, Geyu Liang, and Salar Fattahi
    arXiv preprint, 2025
    arXiv HTML
  3. RANSAC Revisited: An Improved Algorithm for Robust Subspace Recovery under Adversarial and Noisy Corruptions
    Guixian Chen, Jianhao Ma, and Salar Fattahi
    arXiv preprint, 2025
    arXiv HTML
  4. Can Learning Be Explained By Local Optimality In Low-rank Matrix Recovery?
    Jianhao Ma and Salar Fattahi
    Mathematics of Operations Research, 2025
    INFORMS JFIG Paper Competition arXiv HTML
  5. PARQ: Piecewise-Affine Regularized Quantization
    Lisa Jin, Jianhao Ma, Zechun Liu, Andrey Gromov, Aaron Defazio, and Lin Xiao
    In International Conference on Machine Learning, 2025
    HTML

2024

  1. Convergence of Gradient Descent with Small Initialization for Unregularized Matrix Completion
    Jianhao Ma and Salar Fattahi
    In Conference on Learning Theory, 2024
    arXiv HTML

2023

  1. Global Convergence of Sub-gradient Method for Robust Matrix Recovery: Small Initialization, Noisy Measurements, and Over-parameterization
    Jianhao Ma and Salar Fattahi
    Journal of Machine Learning Research, 2023
    arXiv HTML
  2. Behind the Scenes of Gradient Descent: A Trajectory Analysis via Basis Function Decomposition
    Jianhao Ma, Lingjun Guo, and Salar Fattahi
    In International Conference on Learning Representations, 2023
    arXiv HTML

2022

  1. Blessing of Nonconvexity in Deep Linear Models: Depth Flattens the Optimization Landscape Around the True Solution
    Jianhao Ma and Salar Fattahi
    In Advances in Neural Information Processing Systems, 2022
    NeurIPS Spotlight and Katta Murty Prize arXiv HTML
  2. Towards Understanding Generalization via Decomposing Excess Risk Dynamics
    Jiaye Teng*, Jianhao Ma*, and Yang Yuan
    In International Conference on Learning Representations, 2022
    arXiv HTML

2021

  1. Sign-RIP: A Robust Restricted Isometry Property for Low-rank Matrix Recovery
    Jianhao Ma and Salar Fattahi
    In NeurIPS Workshop on Optimization for Machine Learning, 2021
    arXiv HTML