In a finite-dimensional L² projection, an RKHS unit ball becomes a spectral ellipsoid. This schematic connects its covering by ε-balls with kernel eigenvalues—the quantities governing the entropy bounds studied in the paper. Research source
About this paper
Spectral bounds on the ε-entropy of kernel classes, linking approximation complexity to kernel eigenvalues.
The paper studies upper and lower bounds for unit balls of reproducing kernel Hilbert spaces. Eigenvalue decay and convergence of truncated Mercer expansions govern the bounds in different function norms.
Citation
BibTeX
@misc{takhanov2024kernelentropy,
title = {Non-asymptotic spectral bounds on the ε-entropy of kernel classes},
author = {Takhanov, Rustem},
year = {2024},
eprint = {2204.04512},
archivePrefix = {arXiv},
url = {https://arxiv.org/pdf/2204.04512}
}