Speaker
Description
We ask a shot-budget-matched question: at a fixed M shots per timestep, is there a regime where quantum reservoir computing (QRC) beats classical reservoirs sharing the same budget, for a mechanistic reason? A quantum reservoir maps a k-body input observable onto a local Z probe whose shot variance is ~1/M regardless of body count, while a classical shadow estimator pays $3^k/M$; this predicts a QRC edge at intermediate body count and long horizon. We find one clean cell: forecasting a 4-body observable at horizon 20, QRC is predictive (NRMSE 0.986 $\pm$ 0.24) while shot-matched classical-shadow and exact-feature reservoirs both fail (~1.96), over 9 runs (3 Hamiltonians x 3 seeds). Scope is bounded and stated plainly:
one cell, not a trend, and the margin awaits a seed-level interval. We offer it as a proof-of-principle of a measurement-variance mechanism for shot-matched quantum-reservoir advantage.
| Topic | Hybrid quantum-classical approaches for learning and optimization |
|---|