arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2604.23536cs.CV
收藏

$Z^2$-采样:用于扩散模型语义对齐的零成本zigzag轨迹

$Z^2$-Sampling: Zero-Cost Zigzag Trajectories for Semantic Alignment in Diffusion Models

The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州));Griffith University & Data61/CSIRO(格里菲斯大学及Data61/CSIRO)

Haosen Li, Wenshuo Chen, Shaofeng Liang, Lei Wang, Kaishen Yuan, Yutao Yue

AI总结:

本文提出$Z^2$-采样,通过隐式代数坍缩与动态缓存的时间语义代理,实现零成本zigzag轨迹,提升采样效率并保持语义探索。

摘要:

扩散模型在文本对齐生成中取得空前成功,主要得益于分类器自由引导(CFG)。然而,标准CFG仅基于瞬时梯度,忽略了数据流形的内在曲率。最近的方法如zigzag采样(Z-采样)通过多步前向-后向轨迹显式探索曲率,显著提升语义对齐。然而,这些显式轨迹使神经函数评估(NFE)成本增加三倍,并引入无约束截断误差,导致累积漂移。本文理论证明显式zigzag序列可拓扑简化。我们提出隐式Z-采样,严格证明中间状态可通过算子对偶性代数消除,物理上消除流形外近似误差。为推动采样效率至理论下限,我们引入$Z^2$-采样(零成本zigzag采样)。利用概率流ODE的时间一致性,$Z^2$-采样将隐式代数坍缩与动态缓存的时间语义代理结合。这恢复标准2-NFE基线而不牺牲语义探索。通过反向误差分析,我们正式证明这种离散坍缩本质上合成方向导数曲率惩罚。最后,广泛评估显示$Z^2$-采样在结构上打破性能-效率帕累托前沿。我们验证其在不同架构(U-Nets, DiTs)和模态(图像/视频)中的通用性,建立与先进对齐框架(AYS, Diffusion-DPO)的无缝正交性。

英文摘要:

Diffusion models have achieved unprecedented success in text-aligned generation, largely driven by Classifier-Free Guidance (CFG). However, standard CFG operates strictly on instantaneous gradients, omitting the intrinsic curvature of the data manifold. Recent methods like Zigzag-sampling (Z-Sampling) explicitly traverse multi-step forward-backward trajectories to probe this curvature, significantly improving semantic alignment. Yet, these explicit traversals triple the Neural Function Evaluation (NFE) cost and introduce unconstrained truncation errors from off-manifold evaluations, causing cumulative drift from the true marginal distribution. In this paper, we theoretically demonstrate that the explicit zigzag sequence is topologically reducible. We propose Implicit Z-Sampling, rigorously proving that intermediate states can be algebraically annihilated via operator dualities, physically eliminating off-manifold approximation errors. To push sampling efficiency to its theoretical lower bound, we introduce $Z^2$-Sampling (Zero-cost Zigzag Sampling). Exploiting the Probability Flow ODE's temporal coherence, $Z^2$-Sampling couples implicit algebraic collapse with a dynamically cached Temporal Semantic Surrogate. This restores the standard 2-NFE baseline without sacrificing semantic exploration. We formally prove via Backward Error Analysis that this discrete collapse inherently synthesizes a directional derivative curvature penalty. Finally, extensive evaluations demonstrate that $Z^2$-Sampling structurally shatters the performance-efficiency Pareto frontier. We validate its universal applicability across diverse architectures (U-Nets, DiTs) and modalities (image/video), establishing seamless orthogonality with advanced alignment frameworks (AYS, Diffusion-DPO).

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。

↑