Understanding quantum machine learning also requires rethinking generalization

Quantum machine learning models have shown successful generalization performance even when trained with few data. This work demonstrates, through systematic randomization experiments, that traditional approaches to understanding generalization fail to explain the behavior of such quantum models. Experiments reveal that state-of-the-art quantum neural networks accurately fit random states and random labeling of training data. This memorization ability defies current notions of small generalization error, problematizing approaches based on complexity measures. The findings expose a fundamental challenge in the conventional understanding of generalization in quantum machine learning and highlight the need for a paradigm shift.

Elies Gil-Fuster, Jens Eisert, Carlos Bravo-Prieto