Determine converged performance of long-run SimCLR variants

Determine the converged fingerprint-recognition performance of the 200-epoch classical SimCLR model and the original QSSL model using QNet, which were each evaluated only for 15 epochs because of time and hardware constraints.

Background

The paper compares classical SimCLR, QuFeX-based hybrid SimCLR, and QNet-based SimCLR at several training durations. The classical SimCLR and original QSSL configurations were not completed at the longer budgets because of computational constraints, preventing assessment of their converged performance against the longer-running alternatives.

References

Due to time and hardware constraints, the SimCLR-Classic model with 200 epochs, as well as the original version of QSSL (using QNet as the quantum layer) was only run for 15 epochs, so its converged performance is unknown.

— A Width-Matched Comparison of Hybrid Quantum-Classical Self-Supervised Learning for Fingerprint Recognition  (2609.39172 - Edwards et al., 30 Sep 2026) in Section 4.2, “Classical and Hybrid SimCLR with different epoch lengths”