Strategic Classification with Randomised Classifiers

Jack Geary, Henry Gouk · 2025

Read Strategic Classification with Randomised Classifiers by Jack Geary, Henry Gouk (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-2502-01313.txt.

Description

When people can game the features a classifier reads, such as by tweaking a loan application, Jack Geary and Henry Gouk ask whether the classifier should roll dice. Strategic classification studies exactly this contest, and earlier theory assumed the defender always answers the same way for the same input. A randomized classifier instead answers by chance, which may be harder to game.

The analysis shows that, under stated conditions, the best randomized rule can beat the best fixed rule, and it can never do worse. With finite training data, learning the randomized rule carries error bounds much like the fixed case. Its risk shrinks toward the optimum at the usual statistical rate. The authors conclude randomization helps in principle at no real cost.

Catalog record

CREATORJack Geary, Henry Gouk
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-01313

Browse the full library · About the Tower of Babel