Nonequilbrium physics of generative diffusion models
arXiv:2405.11932 · doi:10.1103/PhysRevE.111.014111
Abstract
Generative diffusion models apply the concept of Langevin dynamics in physics to machine leaning, attracting a lot of interests from engineering, statistics and physics, but a complete picture about inherent mechanisms is still lacking. In this paper, we provide a transparent physics analysis of diffusion models, formulating the fluctuation theorem, entropy production, equilibrium measure, and Franz-Parisi potential to understand the dynamic process and intrinsic phase transitions. Our analysis is rooted in a path integral representation of both forward and backward dynamics, and in treating the reverse diffusion generative process as a statistical inference, where the time-dependent state variables serve as quenched disorder akin to that in spin glass theory. Our study thus links stochastic thermodynamics, statistical inference and geometry based analysis together to yield a coherent picture about how the generative diffusion models work.
26 pages, 11 figures, 31 refs
References in corpus (10)
- Stochastic thermodynamics, fluctuation theorems, and molecular machines
- Entropy production along a stochastic trajectory and an integral fluctuation theorem
- Effective potential in glassy systems: theory and simulations
- Origin of the computational hardness for learning with binary synapses
- Rules of calculus in the path integral representation of white noise Langevin equations: the Onsager-Machlup approach
- Dynamical Regimes of Diffusion Models
- Generative diffusion in very large dimensions
- Statistical physics of unsupervised learning with prior knowledge in neural networks
- Minimal model of permutation symmetry in unsupervised learning
- Introduction to dynamical mean-field theory of randomly connected neural networks with bidirectionally correlated couplings