Ascertain end-to-end quantum advantage in machine learning

Ascertain whether end-to-end quantum machine learning pipelines yield robust, practically useful quantum advantages over classical machine learning across broad and realistic families of tasks, beyond specially constructed instances.

Background

The paper surveys quantum machine learning’s promise and caveats: expensive data loading, prior dequantizations of proposed speedups, noisy and weakly structured real-world data, and training obstacles like barren plateaus.

While some rigorous advantages exist for contrived learning settings (e.g., specific generative or classification tasks under hardness assumptions), the authors stress that broad, practically meaningful end-to-end advantages remain to be demonstrated.

References

The potential for useful quantum advantage in end-to-end machine learning tasks remains largely unknown.

— Mind the gaps: The fraught road to quantum advantage  (2510.19928 - Eisert et al., 22 Oct 2025) in Section 4: From near-term quantum heuristics to mature quantum algorithms (tcolorbox)

The debate about whether QML can in general provide a quantum speedup with respect to its classical counterpart is still open, and, while some quantum algorithms for machine learning, such as Fourier transforms, yield an exponential quantum advantage, for others such as finding eigenvectors and eigenvalues, and solving linear equations, a general answer has not been given yet.

— Optical Quantum Computing  (2609.18347 - Hasnaoui et al., 16 Sep 2026) in Section 5.2, “Quantum machine learning”

While our results show a clear advantage for the synthetic tasks we considered, an important open question is for which more realistic tasks QSP-based QCS can also deliver advantages.

— Quantum sensors that compute: quantum computational magnetic-field sensing using a superconducting qubit  (2608.17400 - Sen et al., 18 Aug 2026) in Section Discussion, subsection “Outlook”