Speaker
Description
We show that a quantum ZZ kernel's advantage over classical baselines is governed by one axis: parity
complexity n, the number of features in the XOR rule. On a synthetic XOR benchmark with 22% label noise, we
compare the quantum kernel against an RBF kernel on the identical binary {0, pi} encoding, which isolates
the circuit from the encoding. A clean threshold appears: below n=9 the binary RBF wins, but at n=11
(11 qubits, 10 seeds) all classical methods collapse to chance while the quantum kernel reaches
66.3% $\pm$ 3.2% versus 54.3% $\pm$ 1.1% for the binary RBF, a +12 point margin from the circuit alone, with
7x higher kernel-target alignment. Parity complexity is thus a concrete axis where encoding-independent
quantum-kernel advantage emerges. Scope: synthetic data, exact simulation, at most 11 qubits.
| Topic | Quantum kernel methods and their applications |
|---|