VQIL preprint reports improved ranking of power-grid and gene perturbation responses

Researchers report in a preprint that VQIL, a trainable quantum-walk model, improves predictions of which network components respond most strongly to a disturbance at one node. Across six transmission grids, rank correlation rose from 0.147 for the untrained model to 0.831, outperforming an identically trained random-walk model. VQIL learns parameters attached to individual nodes and connections, combining quantum-walk amplitudes before calculating response scores. Those parameters provide inspectable hypotheses about how disturbances propagate. Across five protein interaction networks, rank correlation rose from 0.171 to 0.620, exceeding the strongest learned classical and neural baselines. The grid tests withheld electrical parameters from the model.