How rotational invariance of common kernels prevents generalization in high dimensions

Authors: Konstantin Donhauser, Mingqi Wu, Fanny Yang

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section we describe our synthetic and real-world experiments to further illustrate our theoretical results and underline the importance of feature selection in high dimensional kernel learning.
Researcher Affiliation Academia 1Department of Computer Science, ETH Z urich. Correspondence to: Konstantin Donhauser <donhausk@ethz.ch>.
Pseudocode No No pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not provide any specific links or explicit statements about the release of open-source code for the described methodology.
Open Datasets Yes In this section we show results on the regression dataset residential housing (RH) with n = 372 and d = 107 to predict sales prices from the UCI website (Dua and Graff, 2017).
Dataset Splits Yes We use a random 80/20 train/test split, and the data is scaled to zero mean and unit variance for each dimension separately.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU, GPU models) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes We primarily use the Laplace kernel with τ = tr(Σd) unless otherwise specified and study two sparse monomials as ground truth functions, f 1 (x) = 2x2 (1) and f 2 (x) = 2x3 (1)... In order to estimate the bias EY ˆf0 f 2 L2(PX) of the minimum norm interpolant we fit noiseless observations and approximate the expected squared error using 10000 i.i.d. test samples.