通过费曼路径积分理解扩散模型
Understanding Diffusion Models by Feynman's Path Integral
AI总结:
本文用费曼路径积分统一表述扩散模型,揭示随机与确定性采样性能差异源于类普朗克常数参数,并借助WKB展开评估负对数似然以解释该差异。
摘要:
基于分数的扩散模型已被证明在图像生成中有效,并已获得广泛应用;然而,导致随机采样方案与确定性采样方案(即概率流常微分方程)之间性能差异的潜在因素仍不清楚。我们引入了一种使用费曼路径积分的新颖扩散模型表述,该表述最初是为量子物理开发的。我们发现这种表述为基于分数的生成模型提供了全面的描述,并展示了后向随机微分方程和损失函数的推导过程。该表述容纳了一个插值参数,连接随机采样方案与确定性采样方案,我们将此参数视为量子物理中普朗克常数的对应物。这一类比使我们能够应用量子物理中成熟的技术——温策尔-克拉默-布里渊(WKB)展开——来评估负对数似然,以评估随机采样方案与确定性采样方案之间的性能差异。
英文摘要:
Score-based diffusion models have proven effective in image generation and have gained widespread usage; however, the underlying factors contributing to the performance disparity between stochastic and deterministic (i.e., the probability flow ODEs) sampling schemes remain unclear. We introduce a novel formulation of diffusion models using Feynman's path integral, which is a formulation originally developed for quantum physics. We find this formulation providing comprehensive descriptions of score-based generative models, and demonstrate the derivation of backward stochastic differential equations and loss functions.The formulation accommodates an interpolating parameter connecting stochastic and deterministic sampling schemes, and we identify this parameter as a counterpart of Planck's constant in quantum physics. This analogy enables us to apply the Wentzel-Kramers-Brillouin (WKB) expansion, a well-established technique in quantum physics, for evaluating the negative log-likelihood to assess the performance disparity between stochastic and deterministic sampling schemes.
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。