Heterogeneous Quantum Federated Learning via Adaptive Circuit Search and Model Aggregation

Shaba Shaon, Atit Pokharel, Alexander Coutras, Avimanyu Sahoo, Dinh C.Nguyen

Keywords: Quantum Federated Learning, Bayesian Optimization, Quantum Architecture Search

TL;DR: Quantum Federated Learning with adaptive circuit search and heterogeneous aggregation for robust performance under data heterogeneity.

Abstract: Quantum Federated Learning (QFL) is an emerging framework for privacy-preserving, collaborative training of quantum neural networks across a network of quantum nodes operating under qubit resource constraints. Although promising, existing QFL approaches enforce a uniform quantum circuit architecture across nodes, failing to account for data heterogeneity and leading to suboptimal global model performance. To tackle these challenges, we propose a BO-QFL framework, which is based on Bayesian optimization to discover node-specific quantum circuit architectures and a novel aggregation rule to unify heterogeneous models at the quantum server. The novel contributions of this paper are twofold: (i) an adaptive circuit architecture search mechanism for heterogeneous quantum nodes, utilizing Bayesian optimization to automatically discover optimal quantum circuit configuration, and (ii) an effective and innovative aggregation strategy that integrates these locally optimized heterogeneous circuits into a unified global model through element-wise logical union. Through rigorous simulations on spatial and temporal datasets, the proposed framework demonstrates a significant improvement in the global model performance over fixed-architecture QFL baselines. Additionally, evaluations in both noisy and ideal quantum environments further substantiate its robustness in realistic quantum settings.

Primary Area: optimization

Submission Number: 22757