---
title: "视频扩散模型中的序列性差距"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.13031v1"
permalink: "https://feed.astralor.com/items/itm_132dd57a5b5d60298c4cac7e"
published: "2026-07-14T17:59:22.000Z"
score: 79
channel: "ai"
tags: ["arXiv","因果链","序列推理","硬球动力学","自回归生成","视频扩散模型"]
---

# 视频扩散模型中的序列性差距

## AI 点评

> 这篇论文做了一件我一直期待有人做的事：把视频扩散模型放在需要逐步因果推理的基准上检验。结果不出所料，模型暴露了硬伤。对于做视频生成、世界模型或物理模拟的研究者，这篇文章是必读的，它点明了当前范式的天花板，并暗示了自回归或混合架构的回归。

## 速览

这篇来自arXiv的论文通过多球硬球动力学实验发现，标准的双向视频扩散模型在因果链变长时性能显著下降，即使增加去噪步数也无法缓解。对照实验表明，问题源于事件间的依赖结构而非视频长度。干预研究显示，增加有效序列计算（如自回归生成、块生成、增加网络深度）能不成比例地提升性能。论文将此现象定义为“序列性差距”：视频扩散模型的去噪循环无法提供可扩展的序列计算能力，并证明了确定性视频预测中，去噪步数不增加网络主干的序列计算量，揭示了视频扩散在序列推理与模拟任务上的结构性障碍。

## 关键点

- 标准双向视频扩散模型在因果链变长时性能下降，增加去噪步数无效。
- 问题根源于事件之间依赖关系导致的序列计算需求，而非视频长度。
- 自回归生成、块生成和网络深度等增加序列计算的方法能有效缓解性能下降。
- 论文从理论上证明去噪步数不增加骨干网络的序列计算量，构成结构性障碍。

## 全文

When one ball strikes another, then another, video models should predict the consequences of each bounce. In controlled experiments on multi-ball hard-sphere dynamics, we find that the performance of standard bidirectional video diffusion degrades as the causal chain lengthens, even when provided more denoising steps. In a length-matched single-ball control, where ball-ball interactions are absent, the degradation largely disappears, isolating dependent-event structure rather than video length as the cause. Across intervention studies, methods that increase effective serial computation improve performance disproportionately, including autoregressive/blockwise generation and architectural depth. We identify this pattern as the seriality gap: a mismatch between tasks requiring growing serial computation and video diffusion models whose denoising loop does not provide scalable serial compute. We then prove that, for deterministic video prediction, denoising steps do not add serial computation beyond the backbone, indicating a structural obstacle for video diffusion on serial reasoning and simulation tasks.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_132dd57a5b5d60298c4cac7e