Papers & Publications

This paper investigates the classical simulation boundaries of Quantum Machine Learning (QML) by applying dequantization techniques to variational quantum circuits. We demonstrate that certain QML models relying on low-rank data representations can be efficiently simulated on classical hardware, challenging assumptions about quantum advantage in machine learning. Our analysis provides formal complexity-theoretic bounds and proposes a taxonomy of QML architectures based on their dequantizability, offering practical guidance on which quantum ML approaches genuinely require quantum computation versus those amenable to classical emulation. The findings have significant implications for resource allocation in near-term quantum computing research and the development of provably quantum-advantaged algorithms.