Nonequilibrium Thermodynamics

Nonequilibrium thermodynamics is about how things naturally flow from high to low - like heat moving from hot to cold regions. Think of it as studying systems actively changing and trying to reach balance rather than systems already at rest.

The math behind Diffusion Models is inspired by the actual physical diffusion processes. Like a drop of ink gradually spreads through water following specific patterns, diffusion models start with random noise and gradually "de-noise" it into clear images or data, following similar mathematical principles. The forward process of a diffusion model mirrors physical diffusion, gradually going from an ordered to a disordered state, and the model learns to reverse it, going from disorder back to order. See Deep Unsupervised Learning Using Nonequilibrium Thermodynamics.