Secure authentication and aggregation in large-scale data-driven systems

Abstract

In the realm of modern technology, large-scale data-driven systems serve as the funda­mental backbone for numerous essential services, ranging from energy management to fire alarms and traffic flow prediction. These systems encompass a network of data sources, intermediate nodes, and a central server. Each source sends its data, through or may not through intermediate nodes, to a server, which processes the data to carry out functions or provide services. This workflow stirs up extensive so­cial debates over security concerns—all involved entities besides the source may abuse its data to infer sensitive contexts or manipulate the data to bring huge threats to both individuals and society. This thesis addresses the critical issue of security in large-scale data-driven systems, with a particular focus on secure authentication and data aggregation. In summary, the contributions of this thesis are: i) approaches to authenticate semantic information in videos and detect two common types of video forgeries: object manipulations (e.g., object removal) and DeepFake manipulations (e.g., face replacement and lip reenactment); ii) a secure data aggregation scheme that realizes group-level security and allows untrusted intermediate nodes, a.k.a., aggrega­tors, to contribute to computation; iii) a novel aggregation mechanism that protects spatio-temporal metadata against the untrusted server and supports efficient batch processing. This thesis advances the current knowledge of data security risks as well as the countermeasures against them and gives a prospect of future works.