Predicting the clean-token posterior and converting it to scores via the known forward kernel (M2S) enforces Bayes-realizable discrete scores and beats SEDD on images and language.
A continuous time framework for discrete denoising models
1 Pith paper cite this work, alongside 4 external citations. Polarity classification is still indexing.
1
Pith paper citing it
4
external citations · OpenAlex
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Mean-to-Score Discrete Diffusion: Posterior-Mean Denoisers for Score Entropy
Predicting the clean-token posterior and converting it to scores via the known forward kernel (M2S) enforces Bayes-realizable discrete scores and beats SEDD on images and language.