Multi-label learning for improving discretely-modulated continuous-variable quantum key distribution

TL;DR

This paper introduces a multi-label learning approach to discretely-modulated CVQKD, enhancing security and performance by allowing larger modulation variance and resisting intercept-resend attacks.

Contribution

It proposes a novel ML-based scheme with a quantum multi-label classifier, improving security and reducing modulation constraints in CVQKD.

Findings

Resists intercept-resend attacks effectively

Enables larger modulation variance in CVQKD

Improves long-distance quantum communication performance

Abstract

Discretely-modulated continuous-variable quantum key distribution (CVQKD) is more suitable for long-distance transmission compared with its Gaussian-modulated CVQKD counterpart. However, its security can only be guaranteed when modulation variance is very small, which limits its further development. To solve this problem, in this work, we propose a novel scheme for discretely-modulated CVQKD using multi-label learning technology, called multi-label learning-based CVQKD (ML-CVQKD). In particular, the proposed scheme divides the whole quantum system into state learning and state prediction. The former is used for training and estimating quantum classifier, and the latter is used for generating final secret key. A quantum multi-label classification (QMLC) algorithm is also designed as an embedded classifier for distinguishing coherent state. Feature extraction for coherent state and related machine learning-based metrics for the quantum classifier are successively suggested. Security analysis shows that QMLC-embedded ML-CVQKD is able to immune intercept-resend attack so that small modulation variance is no longer compulsively required, thereby improving the performance of discretely-modulated CVQKD system.