About Me
Howdy! My name is Yufeng Yang (杨钰峰). I’m a fourth-year PhD student in the Department of Computer Science and Engineering at Texas A&M University, advised by Prof. Yi Zhou. With a background in applied and computational mathematics, I approach computational and algorithmic challenges in large-scale machine learning from first principles. My research on Distributionally Robust Optimization (DRO) has shaped my view that data distributions, structure, and representations should guide model architecture, optimizer design, and training paradigms.
Open to Collaboration! My current interests include:
- Exploring DRO for addressing distribution shifts, with applications to data curation and preference alignment.
- Understanding how the optimization dynamics and generalization of SGD and its preconditioned variants interact with model structure and noisy or heavy-tailed training data, and how optimizer design shapes in-context learning, continual learning, and implicit biases toward particular solutions.
- Developing black-box optimization and reinforcement learning algorithms for agentic workflows, test-time training, and recursive self-improvement.
Interested in collaborating? Email me at yufeng.yang@tamu.edu or connect on WeChat: ynyang94 (please include a brief introduction).
I’m actively seeking AI/ML research or engineering internships starting in Summer or Fall 2027. Here is my CV.
📄First-author Papers
- Distributionally-robust multi-objective optimization [arXiv]
- Nested SGD for Sinkhorn distance-regularized Distributionally Robust Optimization [arXiv] [code] [short version] [poster]
- Adaptive Gradient Normalization and Independent Sampling for (Stochastic) Generalized-Smooth Optimization [arXiv] [code] [slides]