The evidence trick approximates the likelihood score in diffusion inverse problems for exponential family observations by integrating the likelihood against a conjugate variational posterior, enabling Poisson and Binomial posterior sampling with image priors.
Probability and Measure
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
stat.ML 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Diffusion Models for Inverse Problems in the Exponential Family
The evidence trick approximates the likelihood score in diffusion inverse problems for exponential family observations by integrating the likelihood against a conjugate variational posterior, enabling Poisson and Binomial posterior sampling with image priors.