Manifold Diffusion Fields - Learning Generative Models of Continuous Functions on Riemannian Manifolds

Explore a 27-minute conference talk on Manifold Diffusion Fields presented by Ahmed Elhag from Valence Labs. Dive into the innovative approach for learning generative models of continuous functions defined over Riemannian manifolds. Discover how MDF leverages spectral geometry analysis to define an intrinsic coordinate system on manifolds using eigen-functions of the Laplace-Beltrami Operator. Learn about the explicit parametrization method used to represent functions and its invariance to rigid and isometric transformations. Examine empirical results demonstrating MDF's superior performance in capturing function distributions with improved diversity and fidelity compared to existing approaches. Follow the presentation's structure, covering background information, the core concept of Manifold Diffusion Fields, methodology, results, and the transition from meshes to graphs. Engage with the Q&A session at the end to gain further insights into this cutting-edge research in AI for drug discovery.

- Intro + Background
- Manifold Diffusion Fields
- Method
- Results
- From Meshes to Graphs
- Q+A