Prior-Consistent DAPS Improves Diffusion Sampling for Nonlinear Inverse Problems
Original title: Beyond Best-of-$k$: Prior-Consistent DAPS for Ambiguous Nonlinear Inverse Problems
PC-DAPS prunes candidates by prior consistency without ground truth to reduce prior-inconsistent samples on ambiguous nonlinear problems.
Diffusion models serve as priors for inverse problems but struggle with ambiguous nonlinear measurements like phase retrieval where multiple solutions exist. Existing samplers favor measurement-consistent modes unsupported by the prior. Prior-Consistent DAPS maintains multiple candidates within a run and prunes weaker ones based on prior consistency as annealing noise decreases. It reduces prior-inconsistent samples and improves quality on nonlinear problems while preserving results on linear ones.
- Addresses failure where guidance favors non-prior modes
- Prunes candidates according to prior consistency
- Reduces incidence of prior-inconsistent samples
- Improves sample quality on nonlinear inverse problems
- Preserves quality and diversity on linear problems
From the original article
1 min read at arxiv.orgDiffusion models have become strong priors for solving inverse problems. However, nonlinear inverse problems such as phase retrieval remain challenging when multiple measurement-consistent solutions are distinguishable only by the prior; we call such measurements ambiguous. Existing diffusion-based samplers rely on local approximations during reconstruction.
By Merve Kocabas · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.