research
A Diffusion Sampler Trains on Its Own Field Configurations
A target-score identity removes the need for reference samples in a lattice-field experiment.
Summary
A target-score identity removes the need for reference samples in a lattice-field experiment.
The authors train diffusion models from their own generated samples when the unnormalized target density is available. In two-dimensional phi-four theory, selected importance-weighted and fixed-point variants matched training based on Hamiltonian Monte Carlo reference data. The claim is limited to the evaluated field theory.
Why it matters
A target-score identity removes the need for reference samples in a lattice-field experiment.
Limits and context
No additional limitation was separately recorded.
Key claims
A target-score identity removes the need for reference samples in a lattice-field experiment.
Evidence: source-2026-10-14-018
Sources
- arXiv preprint 2610.18863, original v1arXiv / submitting research authors · primary research
Corrections
No corrections have been recorded for this story.