A finite-sample perspective reveals that inexact likelihood approximations cause under- or over-estimation of posterior spread at intermediate timesteps, leading to early-stopping sensitivity, mode weighting errors, and hallucinations even from multimodal priors alone.
Benchmarking Diffusion Annealing-Based Bayesian Inverse Problem Solvers.IEEE Open Journal of Signal Processing, 6:975–991, 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
When, why, and how do diffusion posterior samplers fail? A finite-sample lens
A finite-sample perspective reveals that inexact likelihood approximations cause under- or over-estimation of posterior spread at intermediate timesteps, leading to early-stopping sensitivity, mode weighting errors, and hallucinations even from multimodal priors alone.