New paper gives finite-sample and consistency guarantees for signature kernel ridge regression

chaumian · x · 2026-11-16

A new arXiv paper by Christian Bayer and Jakob Kellermann studies the statistical properties of signature kernel ridge regression. Path signatures are a standard tool for learning functionals on path space, but existing universal approximation results say nothing about how much data is needed. The authors provide finite-sample and consistency guarantees for estimating the conditional expectation E[Y|X] under various moment assumptions, supplemented by numerical experiments. Topics: rough paths, functional data analysis, kernel methods.

Original post →

More from Research

Research channel →