Yadi Cao
Assistant Professor
- Office: SPRK Room 110
- Email: yadi.cao@ucf.edu
- Phone: 407-823-3957
- Google Scholar
- X
BIOGRAPHY
Yadi Cao develops field-ready machine learning for science and engineering: models and agents that hold up not just on clean benchmarks, but under the constraints real research and real hardware impose. His research centers on two directions: generalizable surrogate models that span different geometry, physics, and working conditions without retraining, addressing the data efficiency gap between scientific computing and fields like computer vision or NLP; and cost-aware scientific agents that operate under realistic resource constraints, with benchmarks and post-training methods that account for deployment cost.
Before joining UCF, Cao was a postdoctoral researcher at University of California, San Diego. He received his doctorate in computer science from the University of California, Los Angeles. His work has seen direct adoption in practice: BSMS-GNN enables learning turbomachinery simulations on meshes with over one million nodes and has been adopted by Rolls-Royce, while TGLF-WINN delivers a 48x speedup for fusion transport simulations and is integrated into General Atomics’ tokamak modeling pipeline.
EDUCATION
- Ph.D. in Computer Science – University of California, Los Angeles
- M.S. in Mechanical Engineering – University of British Columbia
- B.S. in Thermal Science and Energy Engineering – University of Science and Technology of China
RESEARCH
- Machine Learning for Science and Engineering
- Neural Surrogate Models
- Cost-Aware Scientific Agents
- Multi-Scale Graph Neural Networks
- In-Context Operator Learning
PUBLICATIONS
- Foam-Agent 2.0: An End-to-End Composable Multi-Agent Framework for Automating CFD Simulation in OpenFOAM. Computer Methods in Applied Mechanics and Engineering, 2025.
- SimulCost: A Cost-Aware Benchmark for Automating Physics Simulations with LLMs. International Conference on Machine Learning (ICML), 2025.
- CFD-LLMBench: A Benchmark Suite for Evaluating Large Language Models in Computational Fluid Dynamics. Journal of Data-centric Machine Learning Research, 2025.
- TGLF-WINN: Data-Efficient Deep Learning Surrogate for Turbulent Transport Modeling in Fusion. Nuclear Fusion, 2025.
- VICON: Vision In-Context Operator Networks for Multi-Physics Fluid Dynamics Prediction. Transactions on Machine Learning Research, 2025.
- Adapting While Learning: Grounding LLMs for Scientific Problems with Intelligent Tool Usage Adaptation. International Conference on Machine Learning (ICML), 2025.
- Physics-Informed Regularization for Domain-Agnostic Dynamical System Modeling. Conference on Neural Information Processing Systems (NeurIPS), 2024.
- Efficient Learning of Mesh-Based Physical Simulation with Bi-Stride Multi-Scale Graph Neural Network. International Conference on Machine Learning (ICML), 2023.
AWARDS
Best Paper Award, NeurIPS DLDE Workshop, 2023