Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Consensus Protocols for Entanglement-Aware Scheduling in Distributed Quantum Neural Networks

Kuan-Cheng Chen, Samuel Yen-Chi Chen, Mahdi Chehimi, Felix Burt, Kin K. Leung

TL;DR

CEAS addresses the challenges of training distributed quantum neural networks over quantum networks by co-designing a fidelity-aware consensus with decoherence-aware entanglement scheduling and quantum authentication for Byzantine resilience. The approach uses fidelity-weighted aggregation grounded in quantum Fisher information $\mathcal{F}_k^{(t)}$ and models entanglement as a perishable resource via a Markov decision process, linking physical-layer parameters to learning dynamics. Theoretical analysis provides convergence intuition under heterogeneous noise, while simulations on a 50-node network demonstrate 10–15 percentage-point gains in accuracy and >90% Bell-pair utilization under coordinated Byzantine attacks, with reduced gradient variance. By integrating cross-layer quantum networking, distributed optimization, and fault-tolerant security, CEAS lays a foundation for scalable, resilient distributed quantum machine learning on a quantum Internet.

Abstract

The realization of distributed quantum neural networks (DQNNs) over quantum internet infrastructures faces fundamental challenges arising from the fragile nature of entanglement and the demanding synchronization requirements of distributed learning. We introduce a Consensus-Entanglement-Aware Scheduling (CEAS) framework that co-designs quantum consensus protocols with adaptive entanglement management to enable robust synchronous training across distributed quantum processors. CEAS integrates fidelity-weighted aggregation, in which parameter updates are weighted by quantum Fisher information to suppress noisy contributions, with decoherence-aware entanglement scheduling that treats Bell pairs as perishable resources subject to exponential decay. The framework incorporates quantum-authenticated Byzantine fault tolerance, ensuring security against malicious nodes while maintaining compatibility with noisy intermediate-scale quantum (NISQ) constraints. Our theoretical analysis establishes convergence guarantees under heterogeneous noise conditions, while numerical simulations demonstrate that CEAS maintains 10-15 percentage points higher accuracy compared to entanglement-oblivious baselines under coordinated Byzantine attacks, achieving 90 percent Bell-pair utilization despite coherence time limitations. This work provides a foundational architecture for scalable distributed quantum machine learning, bridging quantum networking, distributed optimization, and early fault-tolerant quantum computation.

Consensus Protocols for Entanglement-Aware Scheduling in Distributed Quantum Neural Networks

TL;DR

CEAS addresses the challenges of training distributed quantum neural networks over quantum networks by co-designing a fidelity-aware consensus with decoherence-aware entanglement scheduling and quantum authentication for Byzantine resilience. The approach uses fidelity-weighted aggregation grounded in quantum Fisher information and models entanglement as a perishable resource via a Markov decision process, linking physical-layer parameters to learning dynamics. Theoretical analysis provides convergence intuition under heterogeneous noise, while simulations on a 50-node network demonstrate 10–15 percentage-point gains in accuracy and >90% Bell-pair utilization under coordinated Byzantine attacks, with reduced gradient variance. By integrating cross-layer quantum networking, distributed optimization, and fault-tolerant security, CEAS lays a foundation for scalable, resilient distributed quantum machine learning on a quantum Internet.

Abstract

The realization of distributed quantum neural networks (DQNNs) over quantum internet infrastructures faces fundamental challenges arising from the fragile nature of entanglement and the demanding synchronization requirements of distributed learning. We introduce a Consensus-Entanglement-Aware Scheduling (CEAS) framework that co-designs quantum consensus protocols with adaptive entanglement management to enable robust synchronous training across distributed quantum processors. CEAS integrates fidelity-weighted aggregation, in which parameter updates are weighted by quantum Fisher information to suppress noisy contributions, with decoherence-aware entanglement scheduling that treats Bell pairs as perishable resources subject to exponential decay. The framework incorporates quantum-authenticated Byzantine fault tolerance, ensuring security against malicious nodes while maintaining compatibility with noisy intermediate-scale quantum (NISQ) constraints. Our theoretical analysis establishes convergence guarantees under heterogeneous noise conditions, while numerical simulations demonstrate that CEAS maintains 10-15 percentage points higher accuracy compared to entanglement-oblivious baselines under coordinated Byzantine attacks, achieving 90 percent Bell-pair utilization despite coherence time limitations. This work provides a foundational architecture for scalable distributed quantum machine learning, bridging quantum networking, distributed optimization, and early fault-tolerant quantum computation.
Paper Structure (20 sections, 1 theorem, 17 equations, 4 figures)

This paper contains 20 sections, 1 theorem, 17 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Research Directions
  3. Integrated Control Plane for Distributed QML
  4. Robust Consensus with Fidelity Awareness
  5. Decoherence-Aware Entanglement Scheduling
  6. Byzantine Resilience with Quantum Authentication
  7. Hardware-Aware Resource Management
  8. Theoretical Framework
  9. System Model
  10. Distributed Quantum Neural Network
  11. Local Gradient Model
  12. Fidelity-Weighted Consensus Dynamics
  13. Gossip-Based Mixing
  14. Entanglement-Ageing and Scheduling Model
  15. Scheduling as a Markov Decision Process
  16. ...and 5 more sections

Key Result

Proposition 1

Suppose that for all honest nodes $k\in\mathcal{H}$ and iteration $t$, Consider linear aggregators of the form eq:fidelity_weighted_grad over honest nodes only. Among all choices of weights satisfying $\sum_{k\in\mathcal{H}}w_k^{(t)}=1$, the choice minimises the covariance of $\widehat{\boldsymbol{g}}_{\mathrm{agg}}^{(t)}$. In particular, if the scalar fidelity stamp is chosen such that $\phi_k^

Figures (4)

  • Figure 1: Consensus–entanglement–aware training topologies for three different DQNNs schemes.(a) Data-level partition. Each QPU hosts an identical variational circuit $U(\boldsymbol{\theta})\,V(\boldsymbol{\theta})$ and measures local observables. A fidelity-weighted consensus layer (light yellow) aggregates the resulting gradients, distributing high-quality Bell pairs only to the most reliable links. (b) Unitary-level partition. The global variational unitary is factorised into a sequential product $U(\boldsymbol{\theta}) \;\approx\; U_{1}(\boldsymbol{\theta})\,U_{2}(\boldsymbol{\theta}) \,\cdots\, U_{N}(\boldsymbol{\theta})$. Each sub-unitary $U_{i}$ (and its local post-processing block $V_{i}$) is executed on a separate QPU; a layer-wise consensus round synchronises parameter updates before the forward state is teleported to the next node. (c) Circuit-level partition. A large qubit register is partitioned across QPUs. Distributed entangling gates and non-local measurements are orchestrated through the consensus engine, enabling circuits whose width exceeds the native qubit count of any single device.
  • Figure 2: A co-designed architecture for distributed quantum neural network training. The logical layer executes a fidelity-aware consensus protocol, while the physical resource management layer handles entanglement generation and routing across heterogeneous hardware. A cross-layer broker mediates resource allocation based on learning demands.
  • Figure 3: Layered CEAS Control Plane. Illustration of the CEAS architecture integrating a consensus coordination layer (yellow), an entanglement scheduling layer (green), and a cross-layer signalling interface (grey). Distributed QPUs exchange quantum gradient payloads and classical control metadata over authenticated channels to support synchronised model updates under heterogeneous fidelity constraints.
  • Figure 4: Comparison of global accuracy across training rounds for CEAS quantum consensus versus random participant selection, highlighting the stability and accuracy gains from entanglement-aware scheduling.

Theorems & Definitions (5)

  • Definition 1: Fidelity Stamp
  • Proposition 1: Variance Reduction via Fidelity Weighting
  • Definition 2: Entanglement Inventory
  • Definition 3: CEAS Scheduling MDP
  • Definition 4: Byzantine Adversary