A2G-QFL: Adaptive Aggregation with Two Gains in Quantum Federated learning

TL;DR

The paper tackles the instability and performance degradation of federated learning in quantum-enabled, heterogeneous networks by introducing A2G-QFL, a dual-gain aggregation framework with a QoS gain α and a geometry gain β. A2G integrates QoS-weighted trust with curvature-aware manifold corrections, implemented through four phases and a gradient-free, geometry-driven server update. The authors provide convergence guarantees under standard smoothness and bounded variance assumptions, and show that A2G encompasses FedAvg, QoS-aware, and Riemannian FL as special cases. Experiments on a quantum-classical hybrid testbed demonstrate improved stability and higher accuracy under non-IID data and teleportation noise, signaling practical potential for next-generation distributed quantum intelligence.

Abstract

Federated learning (FL) deployed over quantum enabled and heterogeneous classical networks faces significant performance degradation due to uneven client quality, stochastic teleportation fidelity, device instability, and geometric mismatch between local and global models. Classical aggregation rules assume euclidean topology and uniform communication reliability, limiting their suitability for emerging quantum federated systems. This paper introduces A2G (Adaptive Aggregation with Two Gains), a dual gain framework that jointly regulates geometric blending through a geometry gain and modulates client importance using a QoS gain derived from teleportation fidelity, latency, and instability. We develop the A2G update rule, establish convergence guarantees under smoothness and bounded variance assumptions, and show that A2G recovers FedAvg, QoS aware averaging, and manifold based aggregation as special cases. Experiments on a quantum classical hybrid testbed demonstrate improved stability and higher accuracy under heterogeneous and noisy conditions.

Figures (4)

• Figure 1: Quantum Federated Learning: A High Level View of Adaptive Aggregation with Two Gains
• Figure 2: Geometry gain tuning for A2G on two datasets. (a) Breast-Lesions-USG dataset with a non-IID quantity–skew partition: global test accuracy vs. epoch for the FedAvg baseline and A2G with different geometry gains $\beta \in \{0.05, 0.10, 0.30, 0.50, 0.70, 1.0\}$ (with $\alpha=\gamma=\delta=0$). (b) Sklearn Breast Cancer dataset: comparison of QoS-aware A2G with $\alpha=\gamma=\delta=1$ and geometry gains $\beta \in \{0.3, 0.5\}$ against the FedAvg baseline ($\alpha=\gamma=\delta=0,\ \beta=1$).
• Figure 3: QoS trends under geometry–gain tuning on the Breast-Lesions-USG dataset. (a) Mean instability vs. epoch for A2G with geometry gains $\beta \in \{1.0,0.7,0.5,0.3,0.1,0.05\}$ and the FedAvg baseline ($\alpha=\gamma=\delta=0,\ \beta=1$). (b) Mean latency vs. epoch for the same runs. small geometry gains (especially $\beta=0.05$) yield consistently lower instability and latency than both the baseline and the aggressive geometry setting $\beta=1$.
• Figure 4: Impact of A2G under non-IID partitions and different noise regimes on the Breast-Lesions-USG dataset. (a) Global test accuracy vs. epoch for quantity-skewed and label-skewed client partitions, comparing A2G with $\alpha=\gamma=\delta=1,\ \beta=0.05$ against the FedAvg baseline ($\alpha=\gamma=\delta=0,\ \beta=1$) under a medium-noise teleportation channel ($p=0.06$). (b) Global test accuracy vs. epoch for different teleportation noise levels $p \in \{0.01, 0.06, 0.12\}$ using A2G ($\alpha=\gamma=\delta=1,\ \beta=0.05$), contrasted with the FedAvg baseline without teleportation.

Theorems & Definitions (3)

• Lemma 3.6: Single-Step Descent
• Theorem 3.7: Convergence of A2G
• Remark 3.8: Riemannian Case