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 training-free decoding method that uses the bidirectional structure of MDMs to balance immediate uncertainty with the information gained over remaining masked positions. Across reasoning, coding, creative writing, and image generation tasks, Info-Gain Sampler consistently outperforms existing MDM samplers, improving average reasoning accuracy by 2.9--11.6 percentage points and achieving a 62.8% average win rate in creative writing. The code is available at https://github.com/yks23/Information-Gain-Sampler.
@article{arxiv.2602.18176,
title = {Improving Sampling for Masked Diffusion Models via Information Gain},
author = {Kaisen Yang and Jayden Teoh and Kaicheng Yang and Yitong Zhang and Alex Lamb},
journal= {arXiv preprint arXiv:2602.18176},
year = {2026}
}
Comments
https://github.com/yks23/Information-Gain-Sampler Accepted by ICML2026 Accepted by ICML2026