Scientific Frontiers

High-dimensional Generative Foundation Models

This research pillar develops generative foundation models for complex, high-dimensional data — video, time-series, molecular structures — that obey the physical and biological laws of reality, not mere statistical correlation. Guided by this principle of Reliable AI, it reaches ultra-high-resolution imaging and video for medical imaging, materials science, and industrial inspection.

Read more
COD-VAE — class-conditioned 3D shape generation (chair, car) vs VecSet baselines
From “Representing 3D Shapes with 64 Latent Vectors for 3D Diffusion Models” · ICCV 2025

Publications List

High-dimensional Generative Foundation Models