INVESTIGATION OF HYBRID QUANTUM-CLASSICAL ALGORITHMS FOR MACHINE LEARNING

Abstract

Abstract. The architectures of HQML systems leverage the integration of quantum encoding of features with classical optimizations, making them operational within the limits of NISQ devices. In this paper, we provide a systematic review of the literature on HQML between 2018 and 2026, including landmark papers from 2014. Following a PRISMA-inspired approach, 1,247 articles were screened, resulting in 40 key articles for comprehensive assessment. We focus on VQAs, QNNs, and PQCs utilized for classification and optimization tasks in the healthcare and finance industries. Several unresolved technical challenges, such as the barren plateau issue and the burden of data encoding, are addressed. The results show that hybrid approaches exhibit consistent benefits in low-data and class-imbalanced scenarios: increases in accuracy by 0.3% to 2.5% on small medical datasets and equally effective yet data-frugal performance compared to XGBoost for detecting financial fraud. Our study has highlighted five crucial areas that require further research, with the lack of standard benchmarks and learning curve analysis being the most significant.