Pith. sign in

Spectrum Dependent Learning Curves in Kernel Regression and Wide Neural Networks

2 Pith papers cite this work. Polarity classification is still indexing.

2 Pith papers citing it
abstract

We derive analytical expressions for the generalization performance of kernel regression as a function of the number of training samples using theoretical methods from Gaussian processes and statistical physics. Our expressions apply to wide neural networks due to an equivalence between training them and kernel regression with the Neural Tangent Kernel (NTK). By computing the decomposition of the total generalization error due to different spectral components of the kernel, we identify a new spectral principle: as the size of the training set grows, kernel machines and neural networks fit successively higher spectral modes of the target function. When data are sampled from a uniform distribution on a high-dimensional hypersphere, dot product kernels, including NTK, exhibit learning stages where different frequency modes of the target function are learned. We verify our theory with simulations on synthetic data and MNIST dataset.

fields

cs.CV 1 hep-th 1

years

2026 2

representative citing papers

Lectures on Semiclassical Methods for Composite Operators

hep-th · 2026-06-09 · unverdicted · novelty 3.0

Lecture notes develop semiclassical methods to compute large-n scaling dimensions of composite operators in CFTs, recovering known results in free theory and deriving one-loop corrections at the Wilson-Fisher fixed point.

citing papers explorer

Showing 2 of 2 citing papers.

  • SpiS-GAN: Spiral-Modulated Handwriting Synthesis with Star Operation cs.CV · 2026-07-08 · conditional · none · ref 57 · internal anchor

    A GAN with elliptical-spiral feature mixing, star-operation blocks, and Sobel edge loss produces more realistic synthetic handwriting and lowers HTR error rates on English and Vietnamese datasets.

  • Lectures on Semiclassical Methods for Composite Operators hep-th · 2026-06-09 · unverdicted · none · ref 121

    Lecture notes develop semiclassical methods to compute large-n scaling dimensions of composite operators in CFTs, recovering known results in free theory and deriving one-loop corrections at the Wilson-Fisher fixed point.