扩散模型在组合生成任务中难以泛化，尤其当目标分布超出训练范围时。
Catastrophic Compositional Generation: Why Vanilla Diffusion Models Fail to Extrapolate
- 通过理论分析与实验发现，普通扩散模型无法有效处理组合生成任务。
- 在分布外场景下，得分估计误差比推断误差更具破坏性。
- 适合研究生成模型泛化能力或设计更鲁棒的生成方法的读者。
组合生成任务要求条件生成模型在仅训练于部分条件的情况下，生成由源分布组合定义的目标分布样本。本文指出，此类任务对原始条件扩散模型而言往往不可行：我们推测，在某些合理设定下，任何推理阶段技术都无法高效生成目标分布样本。该观点得到理论引导的泛化分析及精心设计的合成与真实数据实验支持。尽管近期方法如Feynman-Kac校正可降低推理阶段近似误差，但我们的结果表明，当目标分布相对于源分布为分布外时，得分估计误差对性能的影响更为严重，凸显了需采用全新方法解决该问题。
原文摘要 · Abstract (English)
The task of compositional generation involves using a conditional generative model, trained only on a subset of the possible conditions, to produce samples from compositionally-defined target distributions such as a geometric combination of the source distributions. In this work, we argue that this task is often infeasible for vanilla conditional diffusion models: we conjecture that no inference-time technique can efficiently produce samples from the target distribution in certain well-motivated settings. This idea is supported by theory-guided generalization arguments and carefully-designed experiments on both synthetic and realistic data. In particular, while recent methods such as Feynman-Kac correction reduce inference-time approximation error, our results show that score estimation error has a more catastrophic effect on performance when the target distribution is out-of-distribution with respect to the sources, highlighting the need for a different approach to this task.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。