Authors: Sarang Vehale, Ruchita Vehale
Abstract: Traditional federated learning (FL) frameworks face critical challenges in privacy, scalability, and auditability when deployed across multiple enterprises with strin- gent regulatory requirements. Quantum-secure protocols such as Quantum Key Distribution (QKD) and post-quantum cryptography can harden communica- tion channels against both classical and emerging quantum attacks. Meanwhile, variational quantum algorithms (VQAs) promise computational speedups for high-dimensional aggregation tasks that become bottlenecks in large-scale FL systems. We propose a hierarchical, multi-tier Quantum-Federated Learning (QFL) architecture in which local enterprises perform classical model training, regional “quantum hubs” execute VQA-accelerated aggregation and anomaly detection, and a global coordinator enforces UN/ISO AI governance via verifiable zero-knowledge proofs (ZKPs). By bounding quantum resource usage to interme- diate nodes and combining QKD on backbone links with lattice-based encryption at the edge, our design achieves near-term implementability, cost-effectiveness, and end-to-end privacy guarantees. Preliminary simulations demonstrate that the proposed scheme reduces communication overhead by over 60% and resists gradient-poisoning attacks with negligible impact on model accuracy. This work lays the foundation for a globally scalable, audit-ready AI governance ecosystem suitable for international deployments
DOI: http://doi.org/10.5281/zenodo.17441935
