Determine whether Piquasso’s ULB advantage persists under repeated sampling

Determine whether Piquasso’s apparent feature-compression advantage on the ULB credit-card fraud dataset persists across repeated sampling runs.

Background

Piquasso achieves unusually strong ULB performance at a very small feature budget, reaching F1 0.874 with five features and outperforming the other tested methods at that budget. The paper notes that repeated runs produced identical rankings in its benchmark, but explicitly leaves unresolved whether the observed advantage is robust to repeated sampling runs. This question concerns the reproducibility of the reported ULB compression result rather than its already-reported benchmark value.

References

Whether Piquasso's ULB advantage survives repeated sampling runs remains to be tested.

— Landscape-Dependent Performance of Photonic Quantum Solvers in QUBO Feature Selection for Financial Risk Detection  (2610.03161 - Sahoo et al., 2 Oct 2026) in Section 6.3, The Small-k Operating Point, subsection “AmEx: No Comparable Sweet Spot”