Changxiao Cai

About

Changxiao Cai is an Assistant Professor in the Department of Industrial and Operations Engineering at the University of Michigan. He previously served as a postdoctoral researcher at the University of Pennsylvania. He earned his PhD in Electrical Engineering from Princeton University in 2021 and his Bachelor of Engineering in Electronic Engineering from Tsinghua University in 2016. His research interests broadly encompass the intersection of statistics, optimization, and machine learning. He focuses on developing provably scalable methods for information extraction from high-dimensional data, aiming to achieve the optimal balance between statistical accuracy and computational efficiency.

Research topics

• Mathematics
• Computer science
• Mathematical optimization
• Algorithm
• Applied mathematics

Selected publications

• Is Q-Learning Minimax Optimal? A Tight Sample Complexity Analysis

  arXiv (Cornell University) · 2021-02-12 · 18 citations

  preprintOpen access

  Q-learning, which seeks to learn the optimal Q-function of a Markov decision process (MDP) in a model-free fashion, lies at the heart of reinforcement learning. When it comes to the synchronous setting (such that independent samples for all state-action pairs are drawn from a generative model in each iteration), substantial progress has been made towards understanding the sample efficiency of Q-learning. Consider a $γ$-discounted infinite-horizon MDP with state space $\mathcal{S}$ and action spa…

  PublisherOA PDFDOI

• Transfer learning for contextual multi-armed bandits

  The Annals of Statistics · 2024-02-01 · 10 citations

  article1st authorCorresponding

  Motivated by a range of applications, we study in this paper the problem of transfer learning for nonparametric contextual multi-armed bandits under the covariate shift model, where we have data collected from source bandits before the start of the target bandit learning. The minimax rate of convergence for the cumulative regret is established and a novel transfer learning algorithm that attains the minimax regret is proposed. The results quantify the contribution of the data from the source dom…

  PublisherDOI

• Subspace Estimation from Unbalanced and Incomplete Data Matrices: $\ell_{2,\infty}$ Statistical Guarantees

  arXiv (Cornell University) · 2019-10-09 · 6 citations

  preprintOpen access1st authorCorresponding

  This paper is concerned with estimating the column space of an unknown low-rank matrix $\boldsymbol{A}^{\star}\in\mathbb{R}^{d_{1}\times d_{2}}$, given noisy and partial observations of its entries. There is no shortage of scenarios where the observations -- while being too noisy to support faithful recovery of the entire matrix -- still convey sufficient information to enable reliable estimation of the column space of interest. This is particularly evident and crucial for the highly unbalanced…

  PublisherOA PDFDOI

• Minimax Estimation of Linear Functions of Eigenvectors in the Face of Small Eigen-Gaps

  arXiv (Cornell University) · 2021-04-07 · 4 citations

  preprintOpen access

  Eigenvector perturbation analysis plays a vital role in various data science applications. A large body of prior works, however, focused on establishing $\ell_{2}$ eigenvector perturbation bounds, which are often highly inadequate in addressing tasks that rely on fine-grained behavior of an eigenvector. This paper makes progress on this by studying the perturbation of linear functions of an unknown eigenvector. Focusing on two fundamental problems -- matrix denoising and principal component anal…

  PublisherOA PDFDOI

• Transfer Learning for Contextual Multi-armed Bandits

  arXiv (Cornell University) · 2022-11-22 · 1 citations

  preprintOpen access1st authorCorresponding

  Motivated by a range of applications, we study in this paper the problem of transfer learning for nonparametric contextual multi-armed bandits under the covariate shift model, where we have data collected on source bandits before the start of the target bandit learning. The minimax rate of convergence for the cumulative regret is established and a novel transfer learning algorithm that attains the minimax regret is proposed. The results quantify the contribution of the data from the source domai…

  PublisherOA PDFDOI

Frequent coauthors

• Yuxin Chen

  16 shared

• H. Vincent Poor

  Princeton University

  12 shared

• Gen Li

  7 shared

• Yuejie Chi

  6 shared

• Yuting Wei

  University of Pennsylvania

  4 shared

• Yuantao Gu

  Tsinghua University

  4 shared

• Gen Li

  Chinese University of Hong Kong

  3 shared

• Sujay Sanghavi

  2 shared

Education

• Ph.D., Electrical Engineering

  Princeton University

  2021

• B.A., Economics

  Tsinghua University

  2016

• B.E., Electronic Engineering

  Tsinghua University

  2016

Similar researchers at University of Michigan

• Mark S. Daskinh-50
• Robert Bordleyh-21
• Albert S. Berahash-16