arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2601.02799cs.LGcs.CL
收藏

分层危险采样:CTMC/DTMC离散扩散和流模型的最小方差事件调度

Systematic Hazard Sampling: Minimal-Variance Inference for Discrete Diffusion and Flow Models

Department of EE KAIST(电子工程系,韩国科学技术院)

Seunghwan Jang, Wonje Jeung, SooJean Han

AI总结:

分层危险采样通过最小化方差提升CTMC/DTMC离散扩散和流模型的样本质量与鲁棒性。

摘要:

均匀噪声离散扩散和流模型(例如D3PM、SEDD、UDLM、DFM)通过迭代地通过多个上下文依赖的替换来非自回归地生成序列。这些模型通常被建模为时间非齐次CTMC/DTMC过程,并在每个离散化步骤中使用独立的伯努利变化决策进行采样。这导致了每个位置跳跃次数的方差呈正增长,从而导致了特征性的不足编辑(残余噪声)和过度编辑(级联替换)故障模式,尤其是在紧缩的离散化预算下会降低样本质量。相比之下,吸收态(掩码开始)模型通过允许每个位置最多跳跃一次来避免这种不稳定性。我们提出了分层危险采样(SHS),这是一种训练自由、可直接使用且超参数自由的推断原则,适用于任何允许停留与替换分解的采样器。SHS将每个token的编辑建模为由累积危险(CTMC)或累积跳跃质量(DTMC)驱动的事件,并通过分层这种累积量来放置事件:每个位置有一个随机相位,当token的累积危险跨越单位间隔的阈值时,该token会被更新。这在保持预期跳跃次数的同时,实现了在无偏整数估计器中的最小可能条件方差(受1/4的限制,对于任何固定的累积质量)。不改变每个跳跃目的地的采样,因此保留了多模态性。在均匀噪声离散扩散语言模型上的实验表明,SHS一致地提高了样本质量。我们进一步表明,SHS在token级黑名单过滤下的鲁棒性得到提升,随着词汇约束的增加,其益处也增加。

英文摘要:

Uniform-noise discrete diffusion and flow models generate sequences non-autoregressively through iterative, context-dependent token replacements. However, these models are typically formulated as time-inhomogeneous continuous- or discrete-time Markov chains (CTMC/DTMC), sampled using independent Bernoulli change decisions per discretization step. This induces Poisson-binomial variance in per-position jump counts that grows with the number of required edits, leading to the common under-editing (residual noise) and over-editing (cascading substitutions) failure modes that degrade sample quality. We identify this sampler-induced variance as an orthogonal source of degradation, distinct from model-side errors and addressable purely at inference time. We propose Systematic Hazard Sampling (SHS), a training-free, drop-in, and hyperparameter-free inference principle for any sampler that admits a stay-vs.-replace decomposition. SHS models per-token edits as events driven by cumulative hazard (CTMC) or jump mass (DTMC) and triggers an edit whenever this quantity exceeds unit-spaced thresholds with a single random phase per position. For any fixed cumulative mass, this preserves the expected jump count while achieving the minimum conditional variance possible among unbiased integer estimators (at most 1/4), without altering per-jump destination sampling. Experiments on four uniform-noise discrete diffusion and flow language models spanning ~110M to ~3B parameters show that SHS consistently improves sample quality across numbers of function evaluations (NFE). Code is available at https://github.com/Jang-seunghwan/Systematic-Hazard-Sampling.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑