English

Loss-Proportional Subsampling for Subsequent ERM

Machine Learning 2013-06-25 v2 Machine Learning

Abstract

We propose a sampling scheme suitable for reducing a data set prior to selecting a hypothesis with minimum empirical risk. The sampling only considers a subset of the ultimate (unknown) hypothesis set, but can nonetheless guarantee that the final excess risk will compare favorably with utilizing the entire original data set. We demonstrate the practical benefits of our approach on a large dataset which we subsample and subsequently fit with boosted trees.

Keywords

Cite

@article{arxiv.1306.1840,
  title  = {Loss-Proportional Subsampling for Subsequent ERM},
  author = {Paul Mineiro and Nikos Karampatziakis},
  journal= {arXiv preprint arXiv:1306.1840},
  year   = {2013}
}

Comments

Appears in the proceedings of the 30th International Conference on Machine Learning

R2 v1 2026-06-22T00:30:11.398Z