On the Protection of Private Information in Machine Learning Systems: Two Recent Approaches

TL;DR

This paper reviews two recent approaches to privacy in machine learning, comparing them with foundational principles from early privacy literature to assess their effectiveness and relevance.

Contribution

It provides a comparative analysis of two recent privacy-preserving methods in machine learning, contextualized within historical privacy principles.

Findings

Both approaches align with classical privacy principles

Older privacy ideas remain relevant for modern ML systems

The review highlights strengths and limitations of current methods

Abstract

The recent, remarkable growth of machine learning has led to intense interest in the privacy of the data on which machine learning relies, and to new techniques for preserving privacy. However, older ideas about privacy may well remain valid and useful. This note reviews two recent works on privacy in the light of the wisdom of some of the early literature, in particular the principles distilled by Saltzer and Schroeder in the 1970s.