Evidential Uncertainty Quantification with Quantum Neural Networks

We introduce a Quantum Evidential Generative Network (QEGN) that assigns quantum neural networks (QNNs) a complementary role as uncertainty quantification heads for classical classifiers. Belief function theory provides a shared semantics: the classical network supplies predictive tendencies expressed as singleton contours, while the quantum network learns higher order relations through evidence assigned to subsets of hypotheses. The proposed method finds a novel role for QNN in machine intelligence, which compacts relational guidance supports learning without a complete target mass distribution. We further extend evidential semantics to general quantum circuits through evidence correction with unitary evolution and give gates an explicit evidential interpretation. Experiments on relation learning, multimodal fusion, and uncertainty quantification support the practical use of the generated evidence.