COMPROMISE: Enhancing Communication Protocols with Machine Learning while Protecting Sensitive Data. Data privacy for communication networks and dynamic databases
Type of activity
AGENCIA ESTATAL DE INVESTIGACION
Funding entity code
Even though the data we voluntarily or involuntarily disclose while browsing the Web or communicating over the Internet through mobile and wireless-network technologies may be of inestimable assistance to scientists and engineers in procuring a sustainable technological progress, the availability of such amounts of information has a crucial downside: they threaten our privacy. To protect the privacy of individuals, current legal frameworks in Europe and other regions limit the collection, processing and sharing of personally identifiable information (PII). The advent of big data has raised the question of how to leverage those PII-data for secondary purposes. It is precisely in this situation where anonymization comes into the picture, as the tool that legitimately allows circumventing the legal restrictions applicable to those data.
In this project we face two complementary goals: (i) Design mechanisms for privacy protection of dynamic data in trusted and untrusted scenarios; and (ii) design network protocols for multihop wireless networks balancing the trade-off between privacy and utility.
On the one hand, we will design anonymization algorithms of dynamic stream data and develop theory and methods of collaborative anonymization. We shall study how to publish protected dynamic data sets for posterior machine learning (ML) analyses. The publication of those dynamic data sets has the advantage that any entity can perform any analyses on the protected data, and that it allows using the protected data, possibly in combination with other data, for secondary purposes. Furthermore, we will design collaborative anonymization algorithms that may empower individuals to protect their data on their own, before handing them to a possibly untrusted data controller.
On the other hand, the privacy mechanisms and anonymization algorithms previously addressed will be applied to communications protocols and services, in order to protect us against different attacks. Firstly, we will focus our research in multihop wireless networks (MWNs) (e.g. vehicular networks, mesh networks), since attacks such as traffic analysis and flow tracing can be easily launched by a malicious adversary due to the open wireless medium. Secondly, we will address the privacy requirements while designing the QoS-aware routing protocols, aiming at improving the trade-off between utility and privacy. Thirdly, we will address emerging services with very exigent QoS requirements and artificial intelligence support that rely on the transmission of parameters from the end nodes to a server, which collects and analyzes data from many nodes to improve the performance of the service. We will design a privacy-aware data collection and processing scheme able to balance privacy and utility using ML techniques to improve the network performance taking advantage of available data and learning capabilities.
Finally, the research results will be integrated into another practical case study, specifically an Android application named MobilitApp that gathers sensor data from citizens smartphones to predict the transportation mode being used. This app will help the metropolitan public service to improve the transport service and to analyze the mobility in the city. We will study the possible loss in accuracy due to the new privacy mechanisms developed in this project and included to protect sensitive sensor data gathered from the smartphones.