Diffusion Models (DMs) and Flow Matching (FM)
This section covers the theoretical and mathematical foundations of diffusion and flow-based generative models.
1. Theoretical Foundation
The various definitions of the diffusion and flow-based generative learning process.
1.1. Diffusion Models (DMs)
1.1.1. Score Matching with Langevin Dynamics (SMLD)
1.1.2. Denoising Diffusion Probabilistic Models (DDPM)
1.1.3. Stochastic Differential Equations (SDEs) Formulation
1.1.4. Graph Diffusion via the System of SDEs (GDSS)
1.1.5. Elucidated Diffusion Models (EDMs)
1.1.6. Cold-Diffusion (CD)
1.1.7. Discrete Denoising Diffusion Probabilistic Models (D3PMs)
1.2. Flow Matching (FM)
2. Samplers
Algorithms for generating samples from the learned generative model.
2.1. Samplers based on Stochastic Differential Equations (SDEs)
2.2. Samplers based on Ordinary Differential Equations (ODEs)
3. Conditioning and Data Consistency
Methods for conditional generation and data consistency in inverse problems.