|
Ph.D. candidate Email: rybindmitry1 [AT] gmail.com |
research interest
I'm a final year Ph.D student at CUHK, lucky to be supervised by Prof. Tom Luo. I work on Machine Learning for Combinatorial Optimization, RL, LLM for Math. I discovered new matrix multiplication algorithms with RL, saving 10% of operations for Causal Attention and XX^T. I am very inspired by scientific contributions done by Google DeepMind and OpenAI, especially: AlphaFold, AlphaTensor, AlphaProof, AlphaEvolve.
biography
I was born in a small manufacturing city Nizhniy Tagil in Russia. At 16, I learned about Russian National Math Olympiad, got top 6 in it and became a Candidate for IMO Team. At 17, I moved to Moscow to study pure math at the world-class department at HSE, and did research on Homological Algebra, Representation Theory, Combinatorics, and Mathematical Physics under supervision of Boris Feigin and Valery Gritsenko.
At 19, I won International Math Olympiad (universty level). At 21, i published first pure math preprint and started PhD at CUHK. There I contributed to various large scale projects e.g. optimization of South Korea 5G Network design. Now I live in Hong Kong / Shenzhen and enjoy their fast-pace high-tech and start-up culture.
education
- 2021 - Present: Ph.D. in Machine Learning, CUHK
- 2017 - 2021: B.Sc. in Mathematics, Higher School of Economics
papers
XX^T Can Be Faster
Dmitry Rybin, Yushun Zhang, Zhi-Quan Luo
Preprint
Finite Horizon Optimization: Framework and Applications
Yushun Zhang, Dmitry Rybin, Zhi-Quan Luo
Preprint
Zeroth-Order Optimization Meets Human Feedback: Provable Learning via Ranking Oracles
Zhiwei Tang, Dmitry Rybin, Tsung-Hui Chang
ICLR 2024
When greedy gives optimal: A unified approach
Dmitry Rybin
Discrete Optimization 2024
Invariant Layers for Graphs with Nodes of Different Types
Dmitry Rybin Ruoyu Sun, Zhi-Quan Luo
Preprint 2023
FFLV polytopes for odd symplectic Lie algebras
Dmitry Rybin
Preprint 2021
blogs
Using GPT-5 to prove new theorem on matrix multiplication
Proving that sequential matrix multiplication is optimal with help of LLMs.
talks
Aug 2025: HPC China
-
Topic: XX^T Can Be Faster
Jul 2025: Peking University / FAI Seminar
-
Topic: Accelerating Structured Matrix Computations
May 2025: Tsinghua University
-
Topic: XX^T Can Be Faster
May 2024: ICLR 2024
-
Topic: Zeroth Order Optimization Meets Human Feedback: Provable Learning via Ranking Oracles
Jul 2023: ICML 2023
-
Topic: Zeroth Order Optimization Meets Human Feedback: Provable Learning via Ranking Oracles
awards
Aug 2021: Presidential PhD Fellowship, CUHK
Aug 2021: Honourable Mention (Top 30/40000), Alibaba Global Mathematics
Aug 2019: Grand First Prize, IMC
Feb 2017: Bronze medal, Russian Team Computer Science Olympiad
Feb 2017: Silver medal, International Zhautykov Olympiad
Apr 2016: Winner, Russian National Math Olympiad