Welcome to Chendong Song's homepage.
I've recently graduated from Zhili College, Tsinghua University with a Bachelor's degree in Mathematics. I am currently a Research Assistant at HKUST IEDA, mentored by Zijie Zhou, and will continue as a PhD student there in 26 Fall.
My research interests lie at the intersection of artificial intelligence and mathematics, spanning both AI for Mathematics (AI4Math) and Mathematics for AI (Math4AI).
In the domain of AI4Math, my past work includes research on the axiomatic systems and symbolic engines for plane geometry—an area inspired by projects such as AlphaGeometry. I have also worked on formalizing geometric problem solving in Lean4, exploring how symbolic reasoning and structured representations can help bridge human-intuitive geometric reasoning. Moreover, I am interested in applying reinforcement learning to train agents capable of solving formal mathematical problems in Lean4.
On the Math4AI side, my research focuses on theoretical optimization of large language model (LLM) inference systems. In particular, I study how probabilistic modeling and combinatorial optimization can be applied to key problems in LLM serving, such as optimal resource provisioning in disaggregated architectures and load balancing in barrier-synchronized decoding systems.
Looking ahead, I am eager to explore algorithm-system codesign for LLM inference acceleration, including parallel decoding strategies, operator-level optimization, and efficient scheduling across heterogeneous hardware. I hope to bridge the gap between theoretical insights and practical system implementation, developing principled yet deployable solutions that push the efficiency frontier of large-scale model serving.
Here is my CV ( in Chinese ).