#machine learning
Feb 2026
Self-Rewarding Sequential Monte Carlo for Masked Diffusion Language Models
This work presents self-rewarding sequential Monte Carlo (SMC), an inference-time scaling algorithm enabling effective sampling of masked diffusion language models (MDLMs), and introduces the trajectory-level confidence as a self-rewarding signal for assigning particle importance weights.
Ziwei Luo, Zi-Qi Jin, Lei Wang et al.
· arXiv.org · 5 citations