Data Free Training of Diffusion Models for Lattice Field Theories
Sampling equilibrium field configurations remains a major computational bottleneck in lattice field theory, motivating the development of generative samplers that can produce independent configurations. Diffusion models are promising for this purpose, but their standard denoising score matching objective requires samples from the target distribution, seemingly reintroducing the
Reading in Plain English mode — toggle below to switch
Sampling equilibrium field configurations remains a major computational bottleneck in lattice field theory, motivating the development of generative samplers that can produce independent configurations. Diffusion models are promising for this purpose, but their standard denoising score matching objective requires samples from the target distribution, seemingly reintroducing the
This paper presents a novel method for training diffusion models without requiring pre-existing samples from the target distribution, addressing a significant bottleneck in scientific computing. For ML engineers, this opens doors to applying generative models in domains where data acquisition is extremely costly or impossible, expanding the utility of diffusion models beyond conventional applications.