Speakers
Description
We employ QCBM to enlarge molecular datasets with artificially generated samples for training SVMs to enhance the classification performance of selected kernel methods. We focus on several datasets including blood-brain barrier permeability prediction, an important task in early-stage drug discovery. On a controlled synthetic benchmark mirroring the topology of AIDS/PROTEINS/NCI1, ER-kernel--guided QCBM augmentation preserves baseline regression quality ($\text{RMSE}=2.40$, $R^{2}=0.77$) and consistently improves on structure-agnostic random augmentation ($\text{RMSE}=2.46$, $R^{2}=0.76$), while the ER-MMD objective decreases monotonically during training. All results use a classical simulation of the Born machine. Our contribution is the coupling of a topology-aware spectral kernel with the MMD training of a Born machine for graph-structured data.
| Topic | Quantum kernel methods and their applications |
|---|---|
| Additional Topic | Quantum algorithms for machine learning applications |