REVIEW 2 cited by
Soft Diffusion: Score Matching for General Corruptions
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Soft Diffusion: Score Matching for General Corruptions
read the original abstract
We define a broader family of corruption processes that generalizes previously known diffusion models. To reverse these general diffusions, we propose a new objective called Soft Score Matching that provably learns the score function for any linear corruption process and yields state of the art results for CelebA. Soft Score Matching incorporates the degradation process in the network. Our new loss trains the model to predict a clean image, \textit{that after corruption}, matches the diffused observation. We show that our objective learns the gradient of the likelihood under suitable regularity conditions for a family of corruption processes. We further develop a principled way to select the corruption levels for general diffusion processes and a novel sampling method that we call Momentum Sampler. We show experimentally that our framework works for general linear corruption processes, such as Gaussian blur and masking. We achieve state-of-the-art FID score $1.85$ on CelebA-64, outperforming all previous linear diffusion models. We also show significant computational benefits compared to vanilla denoising diffusion.
Forward citations
Cited by 2 Pith papers
-
Differencing the Diffusion Trajectory toward Uncertain Components for Time Series Forecasting
DiffDiff rewires diffusion forecasting so corruption gradually emphasizes second-order differences, concentrating generation on history-uncertain parts and improving forecasts on seven benchmarks.
-
Insertion Based Sequence Generation with Learnable Order Dynamics
Learning per-position insertion/unmasking schedules (Kumaraswamy hazard rates) in variable-length discrete flow matching improves validity and quality on molecule generation and graph traversal relative to fixed-sched...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.