Diffusion Models, A Mathematical Perspective
A note on diffusion models, prioritize on the motivations and SODE interpretations.
Notes
Hamiltonian ODEs in the Wasserstein Space
Gradient flows, tangent space, and the Hamiltonian ODEs, in the Wasserstein space of probability measures.
Particle Flow Filters as Generative Diffusion Models
Train a continuous-time generative diffusion model following an interpolation of prior and posterior distributions.