Model the World. Make Sense of It.

【Visualization】

• Scientific Visualization. We develop visualization methods for tensor and vector fields to reveal geometric structures and physical patterns in scientific data. Our research focuses on topology, degenerate structures, and multiscale relationships, combining mathematical analysis with interactive exploration to support the interpretation of complex fields arising from simulations and measurements.

  

• Interactive Visual Analytics. We develop interactive visual analytics systems that integrate visualization, computational analysis, and user interaction to support data exploration and decision-making. Our research includes 3D scene exploration, spatial relationship analysis, and agentic AI applications, emphasizing interface design and interactive workflows that help users investigate complex data, evaluate results, and refine analyses.

  

• Explainable AI. We investigate how language models, vision-language models, and vision-language-action models process tokens and produce responses or actions. Through visual analysis of internal representations and behavioral patterns, we study relationships between token information and model behavior, guiding token selection, compression, and computational optimization while assessing their effects on model performance.

【Modeling】

• Geometric and Scene Modeling. We develop geometric and scene representations from visual observations to reconstruct and understand 3D environments and build digital twins. Our research includes 3D Gaussian Splatting, object-level scene understanding, scene graph construction, and scene updating, integrating geometry, appearance, and semantic relationships to track changes and maintain digital representations of real-world environments.

  

• Physics-Based and Neural Modeling. We combine physical principles, numerical methods, and neural learning to model dynamical systems. Our research includes physics-based simulation, neural PDE solvers, constitutive learning, and inverse problems, incorporating physical and mathematical structure into computational models to support prediction, parameter estimation, and the analysis of complex physical processes.

  

• Perceptual Modeling and Visual Encoding. We model human visual perception and develop visual encoding methods for images and 3D representations. Our research includes image quality assessment, perceptual evaluation of 3D Gaussian Splatting, and hidden-code embedding, connecting perceived quality with visual distortions and balancing perceptual fidelity with reliable message recovery and copy detection.

  

Prospective students must specify which of the research directions above they would like to pursue in their inquiry email.