Geometric Median Matching for Robust k k -Subset Selection from Noisy Data

Anish Acharya, Sujay Sanghavi, Alexandros G. Dimakis, Inderjit S Dhillon · 2025

Read Geometric Median Matching for Robust k k -Subset Selection from Noisy Data by Anish Acharya, Sujay Sanghavi, Alexandros G. Dimakis, Inderjit S Dhillon (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-00564.txt.

Description

Anish Acharya and co-authors propose a robust way to prune training data, in this 2025 paper. Picking a small, telling subset cuts the cost of training large models. Real data sets are noisy, and mean-based picks break under heavy corruption.

Their method matches the subset mean to the geometric median of the full set, a center that resists outliers. Picks are added one by one until the match is close. Theory gives a convergence rate that beats random sampling by a square factor, even under arbitrary corruption. Image tests showed steady wins over rival pruning methods, above all at high corruption and high pruning rates.

Catalog record

CREATORAnish Acharya, Sujay Sanghavi, Alexandros G. Dimakis, Inderjit S Dhillon
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE151 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-00564

Browse the full library · About the Tower of Babel