IEEE Wireless Communications and Networking Conference
DCoRE-QML: Communication-Efficient and Robust Distributed Quantum Machine Learning
Himanshu Sahu, Hari Prabhat Gupta
Abstract
Distributed quantum computing provides scalability while offering advantages such as enhanced noise mitigation and higher fidelity. Leveraging this, distributed quantum machine learning (DQML) can provide better solutions to practical machine learning (ML) tasks, in contrast to classical ML and single-device QML performance. A key objective in DQML is to reduce inter-node communication costs, thereby enhancing efficiency, accuracy, and mitigating communication-induced noise. Existing communication-optimization approaches are generally application-agnostic and do not address the specific demands of QML circuits-especially quantum neural networks (QNNs), which require iterative training, large input sizes, and gradientbased updates. We propose DCoRE-QML, a framework designed for distributed QML that minimizes communication overhead while accounting for network dynamics (link and node failures) to improve robustness. It employs localized data encoding, delayed weight updates, compression techniques, and persistent entanglement purification to improve performance and fidelity. Experimental results indicate that it outperforms generic state-of-the-art solutions for QNN circuits, with higher fidelity.