Bayesian Perspective on Memorization and Reconstruction

TL;DR

This paper presents a Bayesian framework for data reconstruction and proposes a new security definition to prevent reconstruction attacks, offering insights into fingerprinting code attacks and their relation to membership inference.

Contribution

It introduces a Bayesian perspective on data reconstruction, proposes a novel security definition, and reinterprets fingerprinting code attacks as membership inference attacks.

Findings

Proposes a Bayesian paradigm for data reconstruction security.

Reinterprets fingerprinting code attacks as membership inference.

Shows conditions where reconstruction prevention does not imply membership inference vulnerability.

Abstract

We introduce a new Bayesian perspective on the concept of data reconstruction, and leverage this viewpoint to propose a new security definition that, in certain settings, provably prevents reconstruction attacks. We use our paradigm to shed new light on one of the most notorious attacks in the privacy and memorization literature - fingerprinting code attacks (FPC). We argue that these attacks are really a form of membership inference attacks, rather than reconstruction attacks. Furthermore, we show that if the goal is solely to prevent reconstruction (but not membership inference), then in some cases the impossibility results derived from FPC no longer apply.