Overfitting Behaviour of Gaussian Kernel Ridgeless Regression: Varying Bandwidth or Dimensionality

Authors: Marko Medvedev, Gal Vardi, Nati Srebro

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental A Empirical Justification. We plot the dependence of the test error of the Gaussian kernel ridgeless predictor... We ran the experiments on one A6000 GPU.
Researcher Affiliation Academia Marko Medvedev The University of Chicago medvedev@uchicago.edu Gal Vardi Weizmann Institute of Science gal.vardi@weizmann.ac.il Nathan Srebro TTI-Chicago nati@ttic.edu
Pseudocode No The paper does not contain any sections or figures explicitly labeled as 'Pseudocode' or 'Algorithm'.
Open Source Code Yes The code to reproduce these experiments can be found at https://github.com/marko-medvedev/overfitting-behavior-of-gaussian-kernel-ridgeless-regression.
Open Datasets No The paper uses synthetic data generated as 'y = f (x) + ξ where ξ N(0, σ2), f = 10, σ2 is the noise level, and x Unif(Sd 1)', which is not a publicly available or open dataset.
Dataset Splits No The paper describes experiments but does not provide specific training/test/validation dataset splits. It mentions running '100 different runs of the experiment' but no explicit data partitioning.
Hardware Specification Yes We ran the experiments on one A6000 GPU.
Software Dependencies No The paper provides a link to the code for reproduction but does not list specific software dependencies with version numbers.
Experiment Setup Yes Specifically, we consider y = f (x) + ξ where ξ N(0, σ2), f = 10, σ2 is the noise level, and x Unif(Sd 1). We vary the values of d and σ2 and the bandwidth scaling τm as follows: for τm = o(m 1 d 1 ) we take σ2 = 1 and d = 6, for τm = ω(m 1 d 1 ) we take σ2 = 10 and d = 4, and for τm = Θ(m 1 d 1 ) we take σ2 = 10000 and d = 6.