Mingchen Ma (马铭辰)
 |
PhD Student, |
About me
I am a final year PhD student at the Department of Computer Sciences, UW-Madison since 2021. I am very fortunate to be advised by Prof. Christos Tzamos and Prof. Ilias Diakonikolas. I obtained my B.S. in Mathematics in 2020 from the Department of Mathematics at Nanjing University. Before officially starting my Ph.D. study, I spent a year as a visiting student at ITCS, Shanghai University of Finance and Economics.
Research Interests
My research focuses on the algorithmic and theoretical foundations of machine learning and artificial intelligence. I use tools from theoretical computer science, statistics, and optimization to design efficient and reliable learning algorithms or characterize hardness for modern machine learning problems. My current research interests are:
Robust Learning and Information-Computation Tradeoff
Many learning problems can be solved with good sample complexity.
However, designing learning algorithms that achieve both sample efficiency and computational efficiency can be much more challenging and sometimes impossible.
Such a phenomenon frequently happens when label noise or distribution uncertainty is introduced.
I am interested in understanding when designing efficient and robust learning algorithms is possible and when information-computation tradeoff appears.
Foundations and Applications of Interactive Learning
Interactions play a fundamental role in many machine learning problems and are usually used to speed up learning process when data is redundant.
I am interested in understanding what kind of interactions are sufficient and necessary to achieve such improvements.
I am also interested in using the ideas of interactive learning to design better learning algorithm for problems, where interactions are usually not considered.
Principle Understandings of LLMs
Although modern language models have shown amazing capacities, we lack of a good understanding of them in many ways.
Together with my collaborators, we recently focus on hybrid architectures in LLMs.
Using theoretical and empirical tools, we study when hybrid models are better choices than classic pure models and how layer ordering designs affect their performance.
Selected Publications
(Author orders for all papers are alphabetical. Click here for a full list of publications)
Robust Learning of Multi-Index Models under Logconcave Distributions via Weak Partial Learners
with I. Diakonikolas, D. Kane
Proceedings of the 67th Annual Symposium on Foundations of Computer Science (FOCS 2026)Learning Intersections of Two Margin Halfspaces under Factorizable Distributions [conference version]
with I.Diakonikolas, L. Ren, C.Tzamos
Proceedings of the 38th Annual Conference on Learning Theory (COLT 2025)Statistical Query Hardness of Multiclass Linear Classification with Random Classification Noise [arxiv]
with I.Diakonikolas, L. Ren, C.Tzamos
Proceedings of the 42th International Conference on Machine Learning (ICML 2025)
(Selected for Oral Presentation)Active Learning with Simple Questions [arxiv]
with V.Kontonis, C. Tzamos
Proceedings of the 37th Annual Conference on Learning Theory (COLT 2024)Active Learning of General Halfspaces: Label Queries vs Membership Queries [conference version]
with I.Diakonikolas, D.Kane
Advances in Neural Information Processing Systems (NeurIPS 2024)Fast Co-Training under Weak Dependence via Stream-Based Active Learning [conference version]
with I.Diakonikolas, L. Ren, C.Tzamos
Proceedings of the 41th International Conference on Machine Learning (ICML 2024)
(Selected for Oral Presentation)Buying Information for Stochastic Optimization [conference version]
with C. Tzamos
Proceedings of the 40th International Conference on Machine Learning (ICML 2023)
(Selected for Oral Presentation)
Information-Computation Tradeoffs of Robust Learning
Algorithmic Foundations of Interactive Learning
Conference Reviewing
ITCS 2023, STOC 2023, NeurIPS 2024, ICLR 2025, ITCS 2025, ICML 2025, COLT 2025, NeurIPS 2025, ICLR 2026, ICML 2026, NeurIPS 2026, COLT 2026, STOC 2026, FOCS 2026, SODA 2027
Journal Reviewing
Mathematical Programming, Transactions on Machine Learning Research