Skip to content
conftrace_
2013 ICML ICML 2013

Loss-Proportional Subsampling for Subsequent ERM

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.

🚀 Conference Pioneer - ICML 2013
🌉 Interdisciplinary Bridge - Artificial Intelligence and Mathematics & Optimization
🐣 Hot Topic Early Bird - empirical risk minimization

Authors

Paul Mineiro, Nikos Karampatziakis

Topics

Machine Learning > Optimization & Theory > Loss Functions Machine Learning > Application Areas > Efficient Computing Mathematics & Optimization > Optimization > Stochastic Methods Machine Learning > Learning Types > Supervised Learning Machine Learning > Optimization & Theory > Stochastic Methods

Keywords

stochastic optimization empirical risk minimization data reduction excess risk excess risk bound boosted tree loss proportional sampling
Download PDF

More from ICML 2013

All ICML 2013 papers →
Temporal Difference Methods for the Variance of the Reward To Go
\proptoSVM for Learning with Label Proportions
Parameter Learning and Convergent Inference for Dense Random Fields
Scalable Simple Random Sampling and Stratified Sampling
Riemannian Similarity Learning
On Compact Codes for Spatially Pooled Features