Answers are grounded in this paper and cite their sources.
Current approaches to secure gradient aggregation in quantum federated learning typically employ classical or quantum cryptographic techniques to encrypt local gradients prior to transmission to the server via classical or quantum channels. In this process, local gradient information must be transmitted as communication content to take part in aggregation, which increases the potential attack surface for privacy. To address this problem, this paper proposes a quantum secure gradient aggregation scheme based on Ancilla-Driven Quantum Computation (ADQC). The scheme eliminates the direct transmission of local gradient information via either classical or quantum channels. Clients remotely construct single-qubit rotation gates through ADQC, thereby encoding their local gradients into the quantum state held by the server and achieving secure aggregation of local gradients. The proposed scheme includes two protocols for secure gradient aggregation: A serial single-qubit protocol and a parallel multi-qubit protocol. Security analysis demonstrates that both protocols are resilient to external attacks and to attacks originating from servers and clients, thereby achieving privacy protection for local gradients.
No takes yet. Share an insight, caveat, or question.
Shi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: