Quantum Outpost

Algorithm Zoo · Machine learning

Quantum Kernel Methods

Also known as: Quantum SVM, Quantum feature maps

First described: Havlíček et al.; Schuld, Killoran, 2018

Status: Conjectured Maturity: Demonstrated Speedup: None proven on real data

The problem

Classify data using a kernel computed by a quantum feature map.

Map xi → |φ(xi)⟩ with a parameterized quantum circuit; estimate kernel K(xi, xj) = |⟨φ(xi)|φ(xj)⟩|² via overlap measurement; train an SVM with that kernel.

Best classical

RBF kernels, polynomial kernels — extremely well understood and strong.

Quantum complexity

O(D) shots per kernel entry for D-bit precision; O(N²) entries for N data points.

Our verdict

Kernels are a clean way to think about QML but they don't get you out of the QML problem. The Liu-Arunachalam-Temme construction shows a separation exists in principle on a discrete-log-based dataset; no comparable result is known for real datasets. Treat as research, not as production ML.

When to use it

When not to use it

Classical baseline

scikit-learn's RBF kernel SVM trains in minutes on standard benchmarks with accuracy that matches all published quantum-kernel results. Schuld 2021 commentary: 'quantum kernels are kernels' — no exponential separation has been demonstrated on natural data.

Hardware cost

N² overlap circuits — quickly becomes the dominant cost. For 1,000 training points: 500,000 quantum kernel evaluations.

Key papers

Deep-dive tutorials

Last verified: 2026-05-24

Weekly dispatch

Quantum, for people who already code.

One serious tutorial per week, plus the industry moves that actually matter. No hype, no hand-waving.

Free. Unsubscribe anytime. We will never sell your email.