I am a postdoc fellow at the Simons Institute at UC Berkeley hosted by Peter Bartlett and Bin Yu. I am a part of the NSF/Simons Collaboration on the Theoretical Foundations of Deep Learning.
I obtained my Ph.D. in Computer Science at Johns Hopkins University, advised by Vladimir Braverman, and B.S. and M.S. at Peking University. I was selected as a 2023 Rising Star in Data Science by UChicago and UCSD.
Email | CV | arXiv | Google Scholar | Twitter
My research focuses on bridging the gap between theory and practice in machine learning by developing efficient algorithms to solve real-world problems and providing a deep understanding of the underlying theoretical principles.
I work on the theory and algorithms of deep learning, and related topics in algorithms, machine learning, optimization, and statistical learning theory.
Large Stepsize Gradient Descent for Non-Homogeneous Two-Layer Networks: Margin Improvement and Fast Optimization
NeurIPS 2024
Scaling Laws in Linear Regression: Compute, Parameters, and Data
NeurIPS 2024
In-Context Learning of a Linear
Transformer Block: Benefits of the MLP Component and One-Step GD Initialization
NeurIPS 2024
Large Stepsize Gradient Descent for
Logistic Loss: Non-Monotonicity of the Loss Improves Optimization Efficiency
COLT 2024 |
slides |
poster
How Many Pretraining Tasks Are Needed for
In-Context Learning of Linear Regression?
ICLR 2024 (spotlight) |
slides |
poster
Risk Bounds of Accelerated SGD for
Overparameterized Linear Regression
ICLR 2024
Implicit Bias of Gradient Descent for
Logistic Regression at the Edge of Stability
NeurIPS 2023 (spotlight) |
slides |
poster
Private Federated Frequency Estimation:
Adapting to the Hardness of the Instance
NeurIPS 2023 |
slides |
poster
Fixed Design Analysis of
Regularization-Based Continual Learning
CoLLAs 2023
Finite-Sample Analysis of Learning
High-Dimensional Single ReLU Neuron
ICML 2023 |
slides |
poster
The Power and Limitation of Pretraining-Finetuning for Linear Regression under
Covariate Shift
NeurIPS 2022 |
slides |
poster
Risk Bounds of Multi-Pass SGD for Least Squares in the Interpolation Regime
NeurIPS 2022
Last Iterate Risk Bounds of SGD with Decaying Stepsize for Overparameterized Linear
Regression
ICML 2022 (long presentation) |
slides |
poster
Gap-Dependent Unsupervised Exploration for Reinforcement Learning
AISTATS 2022 |
slides |
poster
The Benefits of Implicit Regularization from SGD in Least Squares Problems
NeurIPS 2021
Accommodating Picky Customers: Regret Bound and Exploration Complexity for
Multi-Objective Reinforcement Learning
NeurIPS 2021 |
slides | poster
Lifelong Learning with Sketched Structural Regularization
ACML 2021
Benign Overfitting of Constant-Stepsize SGD for Linear Regression
COLT 2021 (journal version in JMLR 2023) |
slides
Direction Matters: On the Implicit Bias of Stochastic Gradient Descent with
Moderate Learning Rate
ICLR 2021 |
slides |
poster
Obtaining Adjustable Regularization for Free via Iterate Averaging
ICML 2020 |
slides | code
On the Noisy Gradient Descent that Generalizes as SGD
ICML 2020 |
slides |
code
Tangent-Normal Adversarial Regularization for Semi-supervised Learning
CVPR 2019 (oral) |
slides |
poster |
code
The Anisotropic Noise in Stochastic Gradient Descent: Its Behavior of Escaping
from
Minima and Regularization Effects
ICML 2019 |
slides |
poster |
code