Improving Sampling for Masked Diffusion Models via Information Gain
This addresses a fundamental limitation in MDM decoding for researchers and practitioners, offering a novel method to enhance generation quality across tasks like reasoning and creative writing, though it is incremental as it builds on existing MDM frameworks.
The paper tackles the problem of suboptimal sampling in Masked Diffusion Models (MDMs) by proposing the Info-Gain Sampler, which balances immediate uncertainty with information gain over future tokens, resulting in a 3.6% improvement in average accuracy on reasoning tasks and a 63.1% win-rate in creative writing.
Masked Diffusion Models (MDMs) offer greater flexibility in decoding order than autoregressive models but require careful planning to achieve high-quality generation. Existing samplers typically adopt greedy heuristics, prioritizing positions with the highest local certainty to decode at each step. Through failure case analysis, we identify a fundamental limitation of this approach: it neglects the downstream impact of current decoding choices on subsequent steps and fails to minimize cumulative uncertainty. In particular, these methods do not fully exploit the non-causal nature of MDMs, which enables evaluating how a decoding decision reshapes token probabilities/uncertainty across all remaining masked positions. To bridge this gap, we propose the Info-Gain Sampler, a principled decoding framework that balances immediate uncertainty with information gain over future masked tokens. Extensive evaluations across diverse architectures and tasks (reasoning, coding, creative writing, and image generation) demonstrate that Info-Gain Sampler consistently outperforms existing samplers for MDMs. For instance, it achieves a 3.6% improvement in average accuracy on reasoning tasks and a 63.1% win-rate in creative writing. Notably, on reasoning tasks it reduces cumulative uncertainty from 78.4 to 48.6, outperforming the best baseline by a large margin. The code will be available at https://github.com/yks23/Information-Gain-Sampler.