Training-Free Generative Modeling via Kernelized Stochastic Interpolants
This work addresses the computational burden of training neural networks for generative modeling, offering a training-free alternative that is incremental in its approach.
The authors tackled the problem of generative modeling by replacing neural network training with a kernel method based on linear systems, achieving training-free generation and model combination across diverse domains such as financial time series, turbulence, and images.
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 $η_t\in\R^P$ 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 training-free generation and model combination. We demonstrate the approach on financial time series, turbulence, and image generation.