arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.20070cs.LG
收藏

基于核化随机插值的生成建模

Generative Modeling via Kernelized Stochastic Interpolants

CCM, Flatiron Institute, New York, USA(CCM,Flatiron研究所,纽约,美国);Courant Institute of Mathematical Sciences, New York University, New York, USA(数学科学学院,纽约大学,纽约,美国)

Florentin Coeurdoux, Etienne Lempereur, Nathanaël Cuvelle-Magar, Stéphane Mallat, Eric Vanden-Eijnden

AI总结:

提出一种基于核方法的随机插值生成框架,用线性系统替代神经网络训练,通过求解与数据维度无关的线性系统得到生成SDE的漂移项,并处理扩散系数在t=0处的奇异性,适用于多种特征映射,在金融时间序列、湍流和图像生成中验证有效性。

摘要:

我们在随机插值框架内开发了一种用于生成建模的核方法,用线性系统替代神经网络训练。生成SDE的漂移项为$\hat b_t(x) = \nabla\phi(x)^\top\eta_t$,其中$\eta_t \in \mathbb{R}^P$通过求解一个可从数据计算出的$P\times P$系统得到,且$P$与数据维度$d$无关。由于估计不精确,扩散系数$D_t$影响样本质量;从Girsanov定理得到的最优$D_t^*$在$t=0$处发散,但这并不造成困难,我们开发了一种积分器无缝处理该问题。该框架适应多种特征映射:散射变换、预训练生成模型等,无需神经网络训练即可实现生成和模型组合。我们在金融时间序列、湍流和图像生成上展示了该方法。

英文摘要:

We develop a kernel method for generative modeling within the stochastic interpolant framework, replacing neural network training with linear systems. The drift of the generative SDE is $\hat b_t(x) = \nablaϕ(x)^\topη_t$, where $η_t \in \mathbb{R}^P$ solves a $P\times P$ system computable from data, with $P$ independent of the data dimension $d$. Since estimates are inexact, the diffusion coefficient $D_t$ affects sample quality; the optimal $D_t^*$ from Girsanov diverges at $t=0$, but this poses no difficulty and we develop an integrator that handles it seamlessly. The framework accommodates diverse feature maps: scattering transforms, pretrained generative models, etc, enabling generation and model combination without neural network training. We demonstrate the approach on financial time series, turbulence, and image generation.

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。

↑