Skip to content
Source: Quantum Computing Report

⚛️IQM Benchmark Shows 15-Qubit Quantum Kernels Beating Classical ML by 30 Points

Quantum classifiers finally beat classical ones, with caveats

TL;DR

IQM and QbitSoft tested quantum kernels on 15, 30, and 45 qubits against 11 classical models. At 15 qubits, quantum hit 81% accuracy versus 51% for the best classical model. Error mitigation is what separates a real win from noise, so check your baselines before trusting any quantum ML claim.

Quantum machine learning just got a harder test, and it passed one part of it. Researchers at IQM Quantum Computers and QbitSoft ran projected quantum kernels on IQM's Emerald and Garnet superconducting processors and compared them against 11 classical ML models. At 15 qubits with error-mitigated circuits, quantum reached 81% accuracy against 51% for the best classical model. If you're evaluating quantum ML for anything real, the useful part is the referee-challenger framework. It was built to stop models from fitting device-specific hardware noise and inflating their accuracy. The catch: the 15-qubit deep circuit without mitigation only hit 68% versus 64% classical, and that 4-point gap was not statistically significant. Mitigation is doing much of the work. The numbers that matter: a 30-point statistically significant gap at 15 qubits, a 17 to 28-point gap holding at 30 and 45 qubits, and 45-qubit kernels matching a QPU at 72%. The team also reproduced the 45-qubit circuit with a classical matrix product state simulator at bond dimension 128 in under 20 seconds. Next up: 2D quantum dynamics, which is harder to simulate classically and the real test of advantage.

IQM Benchmark Shows 15-Qubit Quantum Kernels Beating Classical ML by 30 Points — Quantum Computing Report

Key Points

1

The 15-qubit projected quantum kernel reached 81% quantum accuracy, compared with 51% for the best of 11 classical ML models, a 30-point statistically significant gap.

Keep reading — it's free

Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.

Also get
or

Already a member? Sign in

quantum-kernelsIQMbenchmarkingerror-mitigationquantum-ml

Comments

Subscribe to join the conversation...

Be the first to comment