LISA mission: quantum neural networks spot all supermassive black hole signals in a test set, without universal quantum advantage

Three researchers compared quantum, hybrid and classical models on simulated Sangria data from the future space gravitational wave observatory.

The LISA space antenna, planned by ESA to detect gravitational waves from space, will have to capture binary mergers of supermassive black holes. Maria-Catalina Isfan, Laurentiu-Ioan Caramete and Ana Caramete wanted to find out if quantum machine learning could help analyse these signals. On the Sangria dataset of the LISA Data Challenge, they opposed several quantum architectures (variational classifiers, quantum kernel methods, data re-encoding classifiers, quantum LSTMs) to classical models: support vector machines, multilayer perceptrons, recurrent networks and LSTMs.

Targeted gains

For signal detection, several quantum models equal or exceed the classical references: the best hybrid architecture correctly identifies all the signals in the test set "blindly". To locate the fusion and estimate the coalescing instant, the standard variational classifier is the most reliable, with the lowest localization errors.

The authors remain cautious: no general quantum advantage appears, but some models offer concrete benefits for specific tasks and deserve a place in future LISA analysis chains, to be weighed against their cost in circuit resources.