Introducing Differential Privacy to the Automotive Domain: Opportunities and Challenges

Abstract

Privacy research is attracting increasingly more attention, especially with the upcoming general data protection regulation (GDPR) which will impose stricter rules on storing and managing personally identifiable information (PII) in Europe. For vehicle manufacturers, gathering data from connected vehicles presents new analytic opportunities, but if the data also contains PII, the data comes at a higher price when it must either be properly de-identified or gathered with contracted consent from the drivers.One option is to establish contracts with every driver, but the more tempting alternative is to simply de-identify data before it is gathered, to avoid handling PII altogether. However, several real-world examples have previously shown cases where re-identification of supposedly anonymized data was possible, and it has also been pointed out that PII has no technical meaning. Additionally, in some cases the manufacturer might want to release statistics either publicly or to an original equipment manufacturer (OEM). Given the challenges with properly de-identifying data, structured methods for performing de-identification should be used, rather than arbitrary removal of attributes believed to be sensitive.A promising research area to help mitigate the re-identification problem is differential privacy, a privacy model that unlike most privacy models gives mathematically rigorous privacy guarantees. Although the research interest is large, the amount of real-world implementations is still small, since understanding differential privacy and being able to implement it correctly is not trivial. Therefore, in this position paper, we set out to answer the questions of how and when to use differential privacy in the automotive industry, in order to bridge the gap between theory and practice. Furthermore, we elaborate on the challenges of using differential privacy in the automotive industry, and conclude with our recommendations for moving forward.

A Differentially Private Hybrid Approach to Traffic Monitoring

“…A comprehensive survey on introduction of differential privacy in the automotive domain is presented by Nelson and Olovsson [5], where they claim that one of the main problems to introduce differential privacy in the automotive domain is maintaining high utility for the analyses. Another important work in this direction is due to Hassan et al [7].…”

“…Among these opportunities, one can highlight investigation of the driver behavior, which helps vehicle manufacturers and insurers to improve and develop new services. Another interesting application is the monitoring of traffic conditions which allows transport departments to manage mobility and improve services [5].…”

“…A vehicle periodically sends beacons collected by sensors to its neighbors, including base stations, which are then sent directly to a TDC. The vehicle sensors collect data such as identification, timestamp, position, speed (direction), acceleration, among other about 7700 signals, some of which are treated as sensitive [4,5].…”

“…Disclosure of this data poses a serious threat to the privacy of contributors, and creates a liability for industry and governments. In Europe, the General Data Protection Regulation (GDPR) imposes stricter rules on the storage and management of personally identifiable information, with non-compliance resulting in severe penalties [5].…”

“…To put it in context, it is worth mentioning that any type of monitoring can lead to a privacy breach through tracking. The main privacy concerns for drivers are disclosure, vehicle tracking and commercial use of personal data [5]. The speed, object of study in this paper, is a vector quantity which has a module (numerical value) and direction.…”