Thursday June 25th, Dr. Antonio Greco will presenta a seminar on”HQ-RCA — From IceCube to IT-Sphere”
The seminar will be online at this link https://meet.google.com/qur-cpwr-afr
Abstract
Banks run continuous IT monitoring that emits a constant stream of alarms. Industrial correlation engines group these alarms but don’t reconstruct the causal chain, so the true root-cause alarm stays hidden and experts must find it by hand.
We borrow an idea from particle physics. In IceCube, rare neutrino events must be separated from an overwhelming background of atmospheric muons. Banking IT is structurally similar: most alarm clusters are background (“log waves”), while rare incidents hide a single root-cause alarm — the analogue of the neutrino vertex. We reuse DynEdge, the graph neural network built for IceCube, to learn over the alarm graph.
On 13 months of real bank data (13k graphs), we propose HQ-RCA: DynEdge’s classical backbone with a compact 4-qubit variational quantum head. It matches the classical baselines on F1 (the standalone DynEdge keeps a small edge on the ranking metrics) and runs on real quantum hardware without error mitigation — a feasible, quantum-assisted workflow, not a decisive quantum advantage.
