Teaching a Transformer to Solve the Rubik's Cube
Six training runs, 165 million parameters, and a hard lesson about the scaling law's limits — curriculum learning, GRPO reinforcement learning, and why ~50% is a structural ceiling.
Recent Posts
Bad Apple!! But It's a DEM Simulation
Recreating the iconic Bad Apple!! music video as a DEM particle simulation — signed distance fields, contact mechanics, and tens of thousands of bouncing particles.
Who Is This Guy Anyway?
A completely accurate and not-at-all exaggerated self-introduction.
Selected Publications
Advanced graph neural network-based surrogate model for granular flows in arbitrarily shaped domains
S. Li, M. Sakai
Chemical Engineering Journal, Vol. 500, 2024
On reduced-order modeling of gas-solid flows using deep learning
S. Li, G. Duan, M. Sakai
Physics of Fluids, Vol. 36, 2024
Projects
GNN Surrogate Models for Granular Flows
Graph neural network-based surrogate models for predicting granular flow behavior in arbitrarily shaped domains.
Deep Learning for Reduced-Order Modeling
Data-driven reduced-order models using POD and deep learning for gas-solid multiphase flow simulations.
About
I am an Assistant Professor at the University of Tokyo. My research focuses on Physics AI, multiphysics simulation, and digital twin technologies. I develop data-driven reduced-order models, graph neural network-based surrogate models, and computational frameworks for large-scale granular flow and multiphase systems.
我是东京大学的助理教授。研究方向为科学智能（Physics AI）、多物理场仿真和数字孪生技术。我致力于开发数据驱动的降阶模型、基于图神经网络的代理模型以及大规模颗粒流和多相流系统的计算框架。
More