Could quantum neural networks trained on LISA data transfer to LIGO-Virgo gravitational wave searches?

Can QNNs trained on LISA data transfer to LIGO-Virgo searches?

The short answer: the archive shows quantum neural networks (QNNs) work for LISA, and transfer learning works for gravitational waves—but there's no direct evidence yet that QNNs specifically transfer between these detector networks.

Here's what we know:

QNNs succeed on LISA data. Researchers have trained QNNs to detect gravitational wave signals in simulated LISA data, achieving classification accuracies above 98% and identifying 5 out of 6 mergers in a blind test dataset [AG-2025.09-333]. This proves the concept works for space-based detectors with their particular noise and signal characteristics.

Transfer learning works for GW detection generally. A classical machine learning model using transfer learning can maintain 99%+ detection accuracy even when detector noise characteristics shift substantially over time, adapting without full retraining [AG-2024.10-284]. This shows that learned features from one noise regime can generalize to another.

But the detectors are fundamentally different. LISA (space-based, low-frequency, long-duration signals from supermassive black holes) and LIGO-Virgo (ground-based, higher frequency, short-duration signals from stellar-mass binaries) observe different source populations with different noise properties. A QNN trained on LISA's signal morphologies—where mergers might take hours to sweep through the band—would see very different time-domain patterns than LIGO's millisecond-scale chirps.

The practical question is whether the learned quantum feature representations are *physics-agnostic enough* to transfer, or whether they encode LISA-specific regularities. The archive doesn't directly address QNN transfer across these detector types—only that transfer learning works for classical networks adapting to changing noise within a single detector.