An optimal diffusion discretization spends equal Fisher-divergence "distance" per step, phi*(s) = Lambda^-1(Lambda s), and the resulting schedule recovers hand-tuned samplers on image benchmarks.
Justification: All the proofs are proven in the supplementary material
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Score-Optimal Diffusion Schedules
An optimal diffusion discretization spends equal Fisher-divergence "distance" per step, phi*(s) = Lambda^-1(Lambda s), and the resulting schedule recovers hand-tuned samplers on image benchmarks.