Structural risk minimization for quantum linear classifiers

Quantum machine learning (QML) models based on parameterized quantum circuits are often highlighted as candidates for computing's near-term “killer application''. However, the understanding of empirical and generalization performance these is still in its infancy. In this paper we study how to balance between training accuracy (also called structural risk minimization) two prominent QML introduced by Havlí?ek et al. \cite{havlivcek:qsvm}, Schuld Killoran \cite{schuld:qsvm}. Firstly, using relationships well understood classical models, prove that model parameters – i.e., dimension sum images Frobenius norm observables used closely control models' complexity therefore performance. Secondly, ideas inspired process tomography, also ability capture correlations sets examples. summary, our results give rise new options minimization models.