Towards Practical Quantum Federated Learning: Enhancing Efficiency and Noise Tolerance

TL;DR

This work introduces two complementary strategies to reduce quantum transmissions: structured parameter reduction based on light-cone feature selection in parametrized quantum circuits, and a Hybrid QFL architecture that dynamically switches from centralized to decentralized aggregation during training.

Abstract

Federated Learning (FL) enables privacy-preserving distributed model training, yet remains vulnerable to gradient inversion and model leakage attacks. Quantum communication has been proposed to provide information-theoretic security for parameter aggregation. However, practical deployment is severely constrained by communication overhead and quantum channel noise. In this work, we present a systematic quantitative study of communication--convergence--noise trade-offs in Quantum Federated Learning (QFL). We introduce two complementary strategies to reduce quantum transmissions: (1) structured parameter reduction based on light-cone feature selection in parametrized quantum circuits, and (2) a Hybrid QFL architecture that dynamically switches from centralized to decentralized aggregation during training. We derive explicit communication cost formulas and show that Hybrid QFL reduces quantum transmissions from $3NMP$ per round to $\{3t + 2(T - t)\}NMP$, achieving substantial savings while preserving near-centralized convergence. We further analyze robustness under depolarizing noise and show that decentralized aggregation is more noise-resilient because it transmits fewer qubits per round. Finally, we evaluate the effectiveness of Steane code-based quantum error correction under high-noise regimes. Our results provide an integrated design framework for communication-efficient and noise-aware QFL, clarifying practical trade-offs necessary for scalable quantum-secure distributed learning.

Figures (12)

• Figure 1: Overview of this work: (a) Quantum Federated Learning utilizing medical data via quantum communications. (b) Two approaches are proposed to reduce the number of quantum transmissions in Federated Learning: the first approach reduces the parameter count by selecting and aggregating only the necessary parameters. (c) The second approach utilizes a Dynamic Network Topology to modify the topology during the training process. (d) Furthermore, we investigate the impact of inherent noise in current quantum communication channels and evaluate whether applying the Steane code can mitigate these noise effects.
• Figure 2: Centralized QFL and Decentralized QFL Protocols: Both protocols begin with GHZ-state generation and parameter encoding. They differ in the redistribution phase. Centralized QFL redistributes the updated global model via quantum teleportation, whereas Decentralized QFL rotates the aggregation role among clients without redistributing a global model.
• Figure 3: Hybrid QFL Protocol: The proposed Hybrid QFL integrates Centralized and Decentralized approaches sequentially. The transition is triggered based on validation performance.
• Figure 4: Overview of Model Architecture: A pre-trained ResNet18 is used as a feature extractor. The extracted features are passed through a classical linear Adapter layer and then into a 6-qubit Quantum Neural Network (QNN). The final classification output is computed as the weighted inner product between the QNN measurement expectations and a trainable vector $\lambda$.
• Figure 5: Light-cone feature selection: The dominant component of the trainable vector $\lambda$ determines the corresponding qubit output. Parameters contributing to this output form a cone-like structure in the quantum circuit, referred to as the light cone. Only parameters within this light cone are selected for aggregation.