On the importance of fundamental properties in quantum-classical machine learning models

Wepresent a systematic study of how quantum circuit design, in particular variational ansatz depth and the choice of quantum feature mapping, affect shybrid quantum classical neural networks for causal classification. The architecture combines a classical convolutional neural network for feature extraction with a parameterized quantum circuit as the quantum layer. We evaluate multiple ansatz depths and nine feature maps. Increasing ansatz repetitions improves generalization and training stability, with benefits saturating at higher depths. Feature mapping choice is more critical since only multi-axis Pauli encodings enable successful learning, while simpler maps lead to underfitting and reduced separability. GRAPHICAL ABSTRACT