Measurement correlations shape practical costs of certifying quantum models
Certification cost of quantum models: measurement correlation, not parameter count
Machine Learning
Summary
Measuring how well quantum computer models learn usually looks at many factors, but often misses how many runs of the circuit are needed to trust these measures. The authors studied how the number of measurements (shots) needed depends more on the correlations in measurement results than on the number of parameters in the model. They found that commonly assumed cubic scaling in parameters only appears in certain cases, and that actual measurement costs can be much lower depending on the setup. They also show ways to optimize measurements to reduce costs and highlight where simple models for error may fail.
What this means in practice
- •For quantum hardware engineers: Estimate realistic measurement runs needed to certify quantum model training in hardware with fixed device connectivity.
- •For quantum algorithm developers: Use optimized measurement weighting strategies to reduce shot costs for evaluating gradients and Fisher information in variational quantum algorithms.
Authors
Pavel Sulimov, Claude Lehmann
Abstract
Reporting the Fisher geometry of a trained variational quantum model is routine; quoting the shot budget that would establish it is not. Certifying an empirical Fisher matrix to relative Frobenius error $\varepsilon$ under coordinate-wise parameter shift costs $Θ(B p^{2} V/(\varepsilon^{2} G))$ circuit executions, where $V$ is the measured readout variance and $G$ the measured squared gradient norm, with uniform allocation optimal in that class. One constant reproduces the cost of two circuit families whose exponents differ by a full power of $p$. The exponent is an identity in how $nV$ and $nG$ scale with the register, holding family by family to $0.001$ across 624 matrix-product-state cells once the finite-$p$ prefactor is removed. The cubic cost is therefore a finite-size window, set by whether the readout light cone grows with the register. A product family to 256 qubits gives $1.966$ (95% CI $1.934$--$1.997$); a brickwork entangler falls from $2.853$ below ten qubits to $1.715$ beyond sixty-four; a blocked entangler gives $1.984$ at a fixed cone width against $3.034$ at a proportional one. Fixed device connectivity fixes the cone, so a cubic budget from a small simulation overestimates a large machine, on top of hardware multipliers $2.07\times$ ($1.41$--$3.02$), $2.38\times$ and $1.91\times$ on ibm_marrakesh, ibm_fez and ibm_kingston. Cost-optimal readout weights cut the measured shot budget by $2.67\times$ ($1.33$--$4.00$) on hardware, flat from four to twelve qubits. A discrepancy model fitted on cheap circuits transfers its mean inside the calibration grid and, at six larger sizes named before the data, does not: nominal 90% intervals cover 36%, and split conformal is the only rung that stays near nominal.