AI RESEARCH
Improving Sampling for Masked Diffusion Models via Information Gain
arXiv CS.CL
•
ArXi:2602.18176v3 Announce Type: replace Masked Diffusion Models (MDMs) enable flexible decoding orders, yet existing samplers remain largely greedy, selecting locally certain tokens without accounting for their downstream effects. We show that this myopia can increase cumulative uncertainty and lead to suboptimal generation. To address this, we propose the **Info-Gain Sampler**, a