IEEE International Symposium on Information Theory (ISIT)
Key-Efficient Decentralized Secure Aggregation with General Security and Collusion Models
Zhou Li, Xiang Zhang, Giuseppe Caire
Abstract
Decentralized secure aggregation (DSA) considers a fully-connected network of K users, each holding a private input. The goal is for every user to compute the sum of all inputs without revealing additional information, even if up to T users collude. Traditional DSA often requires large key sizes to protect all information except the input sum. To reduce key overhead, we study DSA with arbitrary collusion and heterogeneous security constraints, where the inputs of predefined user subsets, called the security set $\mathcal{S}$, must be protected from predefined collusion sets $\mathcal{T}$. For any $\mathcal{S} \in \mathcal{S}$ and $\mathcal{T} \in \mathcal{T}$, we characterize the optimal communication and source key rates. In particular, we determine the minimum number of key bits per input bit required for secure aggregation, which in general reduces to solving a linear program.