Research

Learning-based 3D reconstruction from multi-view images, combining deep learning and geometric reasoning to recover detailed scene structure.

Learning-based novel-view synthesis from sparse input images, combining generalizable Gaussian representations and deep neural rendering to generate photorealistic 3D scene appearances.

Developing generative models for spatiotemporal forecasting and studying the dynamics, stability, and failure modes of iterative AI feedback.

Developing GeoAI methods for analyzing satellite imagery and extracting structured geographic information from large-scale remote sensing data.

Developing case-based reasoning methods that learn useful representations for retrieving and adapting solutions from prior examples.

Developing core deep learning methods for representation learning, model design, and robust visual reasoning.

Surveying diverse topics in AI and computer vision, synthesizing key methods, trends, challenges, and future research directions