← All Research Themes
AI for Science
Overview
Applying AI to scientific discovery, scientific modeling, and complex simulation, with emphasis on differentiable modeling, physics-informed learning, and robust inference.
Key Questions
- How can AI accelerate scientific discovery without sacrificing physical fidelity?
- What role can differentiable simulation play in end-to-end optimization of scientific models?
- How can multimodal AI systems handle complex scientific documents and data?
Our Contributions
- Developed multimodal AI systems for specialized scientific document analysis
- Explored AI methods for differentiable simulation and physics-informed learning
- Studied how scientific models can combine data, structure, and prior knowledge
Recommended Papers
Generates 4D scenes that stay physically consistent across time and materials.
Geometry-aware, anisotropic boundary correction for aerodynamic simulation.
Recent Work
Makes neural operators respect interface physics in multiphase flow simulation.
Embeds physics in graph networks for stable long-horizon multiphysics simulation.
Open Problems
- Bridging the gap between data-driven and physics-based models
- Ensuring interpretability and trustworthiness of AI in scientific applications
- Scaling differentiable simulation to complex real-world systems
Prospective Student Projects
- Differentiable simulation for engineering optimization
- Physics-informed neural networks for dynamic system modeling
- AI-assisted scientific literature mining and knowledge graph construction