The masked diffusion objective decomposes into signal and implicit-regularizer terms, and restricting mask sampling to a signal-rich window improves language model pretraining and fine-tuning at scales up to 8B parameters.
Learning time-scales in two-layers neural networks
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Tuning the Implicit Regularizer of Masked Diffusion Language Models: Enhancing Generalization via Insights from $k$-Parity
The masked diffusion objective decomposes into signal and implicit-regularizer terms, and restricting mask sampling to a signal-rich window improves language model pretraining and fine-tuning at scales up to 8B parameters.