Privacy-Preserving Cloud-Assisted Services

Abstract

In the last decade, there has been a move towards making traditional IT services follow a cloud-assisted services paradigm. This has triggered previously local services to be moved to a cloud-assisted setting to reap the advantages of the cloud-assisted paradigm that can work with simple client-side functionality ("thin clients"). Examples of such services are cloud storage, cloud-assistedmalware checking and "machine learning as a service" (MLaas).

Despite their benefits, these kinds of services put users' privacy at risk since the data stored in the cloud and/or the requests submitted to the cloud may contain sensitive information. On the other hand, unless carefully designed, this service paradigm may nonetheless fail to protect the confidentiality of service providers' business assets (e.g., malware databases or machine learning models) against malicioususers.

This dissertation shows how to leverage cryptographic technologies and trusted execution environments to design cloud-assisted services such that end users can protect their privacy, and if needed, service providers can ensure that their security/privacy requirements are not violated. We provide a general definition for privacy-preserving cloud-assisted services, investigate the privacy issues in three cloud-assisted services: lookup service, prediction service and storage service, and propose solutions on how to make them privacy-preserving.