On the Expressibility and Overfitting of Quantum Circuit Learning

Applying quantum processors to model a high-dimensional function approximator is typical method in machine learning with potential advantage. It conjectured that the unitarity of circuits provides possible regularization avoid overfitting. However, it not clear how interplays expressibility under limitation current Noisy-Intermediate Scale Quantum devices. In this article, we perform simulations and theoretical analysis circuit problem hardware-efficient ansatz. Thorough numerical show generalization error scaling ansatz saturate when depth increases, implying automatic overfitting issue scenario. This observation supported by theory on PAC learnability, which proves VC dimension upper bounded due locality Our study supporting evidence for suppress guidelines performance improvement hardware constraints.