The Computational Wiretap Channel

TL;DR

This paper introduces the computational wiretap channel model, where data sharing via sufficient statistics can be analyzed for privacy and security, approximating the classic wiretap channel under certain conditions.

Contribution

It formalizes the computational wiretap channel and demonstrates its approximation to the Wyner wiretap channel, linking computation and information-theoretic privacy.

Findings

The model captures data sharing and privacy in computational settings.

Under specific conditions, the channel approximates the Wyner wiretap channel.

Provides a theoretical framework for analyzing privacy in computation-based communication.

Abstract

We present the computational wiretap channel: Alice has some data x and wants to share some computation h(x) with Bob. To do this, she sends f(x), where f is some sufficient statistic for h. An eavesdropper, Eve, is interested in computing another function g(x). We show that, under some conditions on f and g, this channel can be approximated, from Eve's point of view, by the classic Wyner wiretap channel.