DUQFL-Prox: Drift-Stable Quantum Federated Learning Framework
The recently introduced DUQFL-Prox framework aims to resolve instability issues in quantum federated learning, enabling distributed clients to train quantum neural networks while keeping their local data private. This method addresses challenges like client drift and uneven performance stemming from diverse data and noisy quantum optimization. By employing deep-unfolded local optimization with adaptive SPSA updates and incorporating a proximal term, DUQFL-Prox ensures that local models remain aligned with the global model. Additionally, a lightweight controller is utilized to learn optimization parameters specific to each step. This approach is particularly suited for privacy-sensitive distributed decision-making systems, including fraud detection and genomic classification.
Key facts
- DUQFL-Prox is a drift-stable quantum federated learning framework.
- It uses deep-unfolded local optimization.
- Each client performs adaptive unfolded SPSA updates.
- A proximal term keeps local model close to global model.
- A lightweight controller learns step-specific optimization parameters.
- Addresses heterogeneous client data and noisy quantum optimization.
- Targets privacy-sensitive distributed decision systems.
- Applications include fraud detection and genomic classification.
Entities
—