arXiv:2604.17954math.DGcs.LG2026-04

复杂归一化流与基尔赫勒-里奇流存在深层几何联系。

Complex normalizing flows can almost be information Kähler-Ricci flows

论文配图:Complex normalizing flows can almost be information Kähler-Ricci flows
图 1 · 摘自论文原文
  • 通过威廷格雅可比行列式对数行列式,建立数据流形变换与曲率的关联。
  • 在连续极限下,对数似然匹配费舍尔度量,逼近基尔赫勒-里奇流。
  • 为复归一化流提供几何解释,适合几何深度学习研究者阅读。

我们建立了复归一化流与近似基尔赫勒-里奇流之间的联系。复归一化流描述了在复流形实化空间上,从初始分布到目标分布的变量变换,其核心依赖于一组威廷格雅可比矩阵的对数行列式。而基尔赫勒流形的里奇曲率正是体积形式局部密度对数的二阶混合威廷格偏导数。因此,通过微分运算可发现:归一化流中的对数行列式项恰好对应于里奇曲率项。此外,在参数的贝叶斯视角和增强雅可比框架下,对数密度与空间信息度量相关,当趋于连续极限时,对数似然演化为费舍尔度量,更精确地表现为一个基尔赫勒交叉熵黑塞矩阵。这使得整体演化接近基尔赫勒-里奇流,仅差时间导数与期望项,或等价于平均值型基尔赫勒-爱因斯坦流。该框架进一步揭示了复归一化流的统计行为与几何特征之间的内在一致性。

原文摘要 · Abstract (English)

We develop interconnections between the complex normalizing flow for data drawn from Borel probability measures on the twofold realification of the complex manifold and a nonlinear flow nearly Kähler-Ricci. The complex normalizing flow relates the initial and target realified densities under the complex change of variables, necessitating the log determinant of the ensemble of Wirtinger Jacobians. The Ricci curvature of a Kähler manifold is the second order mixed Wirtinger partial derivative of the log of the local density of the volume form. Therefore, we reconcile these two facts by drawing forth the connection that the log determinant used in the complex normalizing flow matches a Ricci curvature term under differentiation and conditions. The log density under the normalizing flow is kindred to a spatial information metric under an augmented Jacobian and a Bayesian perspective to the parameter, thus under the continuum limit the log likelihood matches a Fisher metric, or more closely a Kähler cross-entropy Hessian. This recovers a Kähler-Ricci flow variation up to a time derivative and expectation, or an average-valued Kähler-Einstein flow. Using this framework, we establish other relevant results, attempting to bridge the statistical and ordinary behaviors of the complex normalizing flow to the geometric features of our derived Kähler flow.

归一化流基尔赫勒几何概率流

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。