Evaluating Quantum Advantage in Noisy Machine Learning: A Survey of Hardware Degradation, Mitigation Overhead, and Algorithmic Resilienc

Abstract

Variational Quantum Algorithms (VQAs) and Quantum Neural Networks (QNNs) represent the primary frontier for near-term quantum advantage. However, physical noise channels on Noisy Intermediate-Scale Quantum (NISQ) devices severely degrade classification accuracy and deform optimization landscapes. This survey synthesizes findings across recent empirical and theoretical studies, mapping the transition from raw phenomenological noise vulnerability to active error mitigation and passive algorithmic co-design. We examine asymmetric noise sensitivity, the sampling overhead bottleneck of active mitigation, and the threshold limitations of logical stabilizer codes. Finally, we outline a consensus roadmap highlighting hybrid, sensitivity-pruned, low-depth ensembles as the optimal strategy for NISQ-era machine learning