Browse Teach Curate

Multi-Objective Optimization for the Privacy-Utility Tradeoff in Data Anonymization

An anonymization method that jointly optimizes data utility and privacy as competing objectives - minimizing information loss, computed distinctly for numerical and categorical attributes, while maximizing the entropy of sensitive values within each group - rather than minimizing loss subject to fixed privacy constraints, yielding tunable privacy-utility solutions that preserve machine-learning performance.

Rosie Text

Multi-Objective Optimization-Based Anonymization of Structured Data for Machine Learning

This paper formulates data anonymization - transforming a dataset so individuals cannot be re-identified before it is shared with external analysts - as a multi-objective optimization problem that si…