Girard, M Fabien (2024) Matrix profile & Privacy PRE - Research Project, ENSTA.

[img]
Preview
PDF
1053Kb

Abstract

The Matrix Profile is a recent statistical tool that has attracted a lot of interest among data mining researchers. It can be used, quite spectacularly, to detect anomalies, patterns, or even similarities in a time series. So, theoretically, it enables us to maintain a certain confidentiality over data while still accessing a certain amount of exploitable information. The question is: does the Matrix Profile provide so much information that it compromises the confidentiality of the time series from which it is derived? In this article, we will look at this last point by building, step by step, an attack consisting of inverting the Matrix Profile to retrieve the time series from which it was derived. Our main motivation is to make potential users of this tool aware that it represents much more than a simple average, which could be sent to a data analyst without the risk of saying too much. The attack we’ve built is therefore intended to be as general as possible, to cover all use cases, even if we end up refining it on an ad hoc basis for a certain type of sensitive data.

Item Type:Thesis (PRE - Research Project)
Uncontrolled Keywords:Matrix-Profile ; Privacy ; Attack
Subjects:Information and Communication Sciences and Technologies
Mathematics and Applications
ID Code:10114
Deposited By:Fabien GIRARD
Deposited On:09 sept. 2024 17:07
Dernière modification:09 sept. 2024 17:07

Repository Staff Only: item control page