Welcome to the beautiful world of SBI, with this terrific piece of work by Pippa Cole. Here we’re looking at extreme mass-ratio inspirals (EMRIs), that is, a small black hole orbiting a big black hole, which will be (one day) detected by LISA. These signals are nasty (long and of a very complicated morphology). We’re trying something new here – a deep learning called “truncated marginal neural ratio estimation” that does not even require writing down the likelihood of the problem. Just simulate all you can. The answer, this thing is great for narrowing down the parameter space where EMRIs will be, kind of like searches do with current gravitational-wave data, but in a very different way.

Philippa S. Cole, James Alvey, Lorenzo Speri, Christoph Weniger, Uddipta Bhardwaj, Davide Gerosa, Gianfranco Bertone.
arXiv:2505.16795 [gr-qc].