Quantum machine learning is an emerging discipline that aspires to combine the best of quantum computing and machine learning worlds. Many quantum-based models have been proposed over the recent years that showed promising results in their specific domains. Yet little is known about how to design an effective hybrid classical-quantum model. This work is shedding some light on this topic by conducting an extensive experiment to show how including quantum information into the classical model for anomaly detection affects its overall performance. The classical part of the models has a typical autoencoder design that remains unchanged throughout the experiment, while "Ansätze" of parametrized quantum circuits (PQC) vary. The experiment reveals correlations between novel PQC expressivity descriptors and F1 score that was calculated on a real-life dataset. Based on these results, a novel quantum variational advesarial autoencoder model is proposed.

Keywords: Quantum Machine Learning, Autoencoders, Anomaly Detection