QViT-Exo: Uncertainty-Aware Exoplanet Transit Classification with Quantum Attention and Conformal Prediction

Approximately 50% of TESS transit candidates are false positives arising from eclipsing binaries, instrumental artefacts, or background contamination. Existing classifiers such as ExoMiner++ achieve high recall but provide neither calibrated uncertainty estimates nor statistically validated interpretability. This technical note presents QViT-Exo, a hybrid quantum-classical Vision Transformer that addresses both limitations simultaneously. The model processes dual-channel 2D representations of phase-folded light curves fused with auxiliary diagnostic features through a ViT-B/16 backbone enhanced with Quantum Orthogonal Neural Network (QONN) attention layers. Uncertainty quantification is provided by Adaptive Quantum Conformal Prediction (AQCP), which produces prediction sets with provable finite-sample marginal coverage guarantees. On the Kepler DR25 catalog, QViT-Exo achieves 91.2% recall with a 56% false-positive rate reduction via selective abstention. Full code is available at https://github.com/Shy4n7/QViT-Exo