Generative Modeling
Diffusion, flow matching, stochastic processes, discrete generation, and related sampling methods.
- A unified derivation of diffusion and flow matching through conditional probability paths, marginalization identities, and simulation-free regression.
- How continuous-time Markov chains transport categorical probability, how their rates become learnable, and how generator matching extends across modalities.
- How ODEs and SDEs transport probability, why scores appear in reverse-time diffusion, and how probability-flow ODEs match SDE marginals.
- GFlowNets from a probabilistic-ML perspective: reward-proportional sampling, training objectives, and connections to MaxEnt RL and variational inference.
- How path measures connect Jarzynski's equality, free-energy estimation, annealed importance sampling, diffusion models, and GFlowNets.
- Three routes to the Fokker-Planck equation—physical intuition, heuristic discretization, and a rigorous derivation with Itô calculus.