最陡引导：流与扩散模型推理时对齐的一种实用且规范的方法
Steepest Guidance: A Practical and Principled Approach to Inference-Time Alignment of Flow and Diffusion-based Models
浏览论文内容
中文总结 AI 辅助
针对流与扩散模型推理时对齐的挑战，提出基于局部改进最大化原理的最陡引导框架，理论分析并实验验证其有效性。
中文摘要 AI 辅助
流与扩散模型的推理时对齐对于实现灵活的生成建模至关重要。理论上，Doob的$h$-变换为该问题提供了优雅的解决方案，且大多数现有方法均基于此原理。然而，在实践中，推理时估计由Doob的$h$-变换导出的最优引导具有挑战性。为解决此问题，我们将推理时对齐视为概率测度空间中的序贯优化问题，并提出一种名为*最陡引导*的新框架，其基于目标局部改进最大化的原理。我们对该方法进行了理论分析，并通过大量实验证明了其有效性。
英文摘要
Inference-time alignment of flow and diffusion-based models is critical for achieving flexible generative modeling. Theoretically, Doob's $h$-transform provides an elegant solution to this problem, and most existing methods are based on this principle. However, in practice, estimating the optimal guidance derived from Doob's $h$-transform at inference time is challenging. To deal with this issue, we regard inference-time alignment as a sequential optimization problem in the space of probability measures and propose a novel framework called *Steepest Guidance*, based on the principle of maximizing local improvement in the objective. We provide a theoretical analysis of the proposed method and demonstrate its effectiveness through extensive experiments.
发表机构
- LY Corporation（LY公司）
- The University of Tokyo（东京大学）
- RIKEN AIP（理化学研究所人工智能研究中心）
机构由 AI 辅助整理，请以论文原文为准。
