|
Yuhan YeI am a PhD student at the MIT Center for Computational Science and Engineering (CCSE), where I fortunately work with Professors Asu Özdağlar and Saurabh Amin. Before joining MIT, I earned a B.S. in applied and computational mathematics from Peking University in 2025. News: I was selected as a 2026–2027 MIT Schwarzman College of Computing PhD Fellow. |
Research
My research focuses on algorithm design and theoretical analysis in optimization, learning, and EconCS.
A central question in my work is what information, and how much of it, is truly needed to make optimal decisions. To address this, I develop learning algorithms that preserve decision quality while minimizing both the amount of information required and the size of the resulting optimization models. This work lies along the lines of beyond-worst-case analysis and data-driven algorithm design, connecting ideas from mechanism design, optimization, learning theory, and computational complexity.
Publications
-
Silver Rate Is (Almost) Optimal for Gradient Descent
-
Improved Gradient Descent Lower Bounds Beyond Nesterov
arXiv preprint, 2026. [arxiv]
-
The Geometry of Linear Program Compression: An Exact Characterization and Learning Algorithm
arXiv preprint, 2026. [arxiv]
-
Parameterized Complexity of Stationarity Testing for Piecewise-Affine Functions and Shallow CNN Losses
arXiv preprint, 2026. [arxiv]
-
Learning Decision-Sufficient Representations for Linear Optimization
, Saurabh Amin, Asuman Özdağlar
COLT 2026. [arxiv] [conference] [slides]
STOC 2026 workshop: Machine Learning for Algorithms. [workshop] [slides]
-
An Online Adaptive Sampling Algorithm for Stochastic Difference-of-convex Optimization with Time-varying Distributions
ICML 2025, oral. [pdf] [conference] [poster]
1. Theoretical analysis and efficient algorithms for optimization
-
Silver Rate Is (Almost) Optimal for Gradient Descent
-
Improved Gradient Descent Lower Bounds Beyond Nesterov
arXiv preprint, 2026. [arxiv]
-
An Online Adaptive Sampling Algorithm for Stochastic Difference-of-convex Optimization with Time-varying Distributions
ICML 2025, oral. [pdf] [conference] [poster]
-
Parameterized Complexity of Stationarity Testing for Piecewise-Affine Functions and Shallow CNN Losses
arXiv preprint, 2026. [arxiv]
2. Information design for optimization
-
Learning Decision-Sufficient Representations for Linear Optimization
, Saurabh Amin, Asuman Özdağlar
COLT 2026. [arxiv] [conference] [slides]
STOC 2026 workshop: Machine Learning for Algorithms. [workshop] [slides]
-
The Geometry of Linear Program Compression: An Exact Characterization and Learning Algorithm
arXiv preprint, 2026. [arxiv]
Contact
Feel free to reach out to me at:
- Email: yyh03@mit.edu
Miscellaneous
I like playing chess, while I am struggling to improve my ELO on chess.com. I also play other strategy games.