Graph Algorithms over Homomorphic Encryption for Data Cooperatives
Mark Dockendorf, Ram Dantu, John Long · 2022
There are a number of issues facing us today regarding data privacy and data access. While most individuals and organizations would like to use their data in ways that benefit themselves and others, current solutions leave much to be desired from a privacy and security standpoint. Alex Pentland, et al. proposed using data as a form of capital and designed data cooperatives as a means to share and extract value from data. Their work on data cooperatives is akin to designing a data model, in that it was abstract, high-level, and focused on the capabilities it brings to bear (in-line with their research focus). This work proposes a more concrete way of building a data cooperative and addresses some of the real-world pain points of implementing a data cooperative using homomorphic encryption (HE) to secure data. This work was conducted focusing on pooling data between untrusting partners for cooperative cyber-defense. First, there is a design for different internal components for building a data cooperative. Next, this work implements a federated graph storage technique for a graph database that works symbiotically with the novel HE query watchdog. This watchdog enforces k-anonymity (via rules derived from the privacy policy) on top of HE and ensures no query that violated k-anonymity along any step will ever have its result delivered. This work then demonstrates numerous commonly used graph algorithms with modified implementations to work over HE. Finally, given the burden of HE, this work explores methods to reduce the burden of HE on the data cooperative, while still ensuring k-anonymity for all participants and security concerns are addressed.