This work provides a comprehensive study of generalization performance in Quantum Machine Learning (QML) with limited training data. The authors show that generalization error scales at worst as T/N (where T is the number of trainable gates and N is the number of training data points), improving to K/N when only K << T gates have significantly changed during optimization. This implies significant speedups for applications like unitary compiling. The study also demonstrates the effectiveness of QML for quantum state classification and highlights potential applications in quantum error correction and dynamical simulation.

Matthias C. Caro, Hsin-Yuan Huang, M. Cerezo, Kunal Sharma, Andrew Sornborger, Lukasz Cincio, Patrick J. Coles