VISTA-SMC：离散扩散中加速序列蒙特卡洛引导的值信息调度裁剪
原标题：Value-Informed Schedule Trimming for Accelerated Sequential Monte Carlo Guidance in Discrete Diffusion
AI 摘要
该论文提出 VISTA-SMC，一种无需训练、即插即用的方法，用于在离散扩散模型的序列蒙特卡洛（SMC）引导中自动选择预算受限的引导时间步。现有稀疏引导方案多为启发式固定调度，难以适应不同任务与奖励景观。VISTA 将稀疏 SMC 引导建模为对完整引导概率路径的近似问题，通过可估计的上界代理目标并利用动态规划精确求解调度。在 Binary MNIST、有毒文本和 DNA 序列生成任务上，VISTA 在相同预算下多数设置优于手工设计的稀疏基线，并相对完整引导实现大幅加速。
正文节选
1 Introduction Discrete diffusion models [2] have emerged as a powerful framework for generating structured objects such as text, biological sequences, and discrete images. In these domains, practitioners often prioritize not only the likelihood under a pretrained model but also alignment with external objectives, such as safety in language, biological activity in DNA, or constraint satisfaction. Consequently, a primary goal is to sample from a reward-tilted distribution that steers the pretrain