[polyphaser]

There's no fancy machine learning needed to detect some "bright" LIGO signals (some of the first blackhole-blackhole mergers). Given a set of template signals, a matched filter tries to find the one that best matches the noisy signal your instrument recorded. In order for a MF to work, what you really need is a good understanding of the noise in your instruments observations -- and there're very few people in the world better at that than LIGO folks. LIGO spent almost 20 years in construction and R&D, and almost 30 in planning. When you hear stories about how they can detect trucks miles away, and detect the waves crashing on the coast, it's possible because of scores of PhD students' who spent years characterizing each and every component that affects LIGO's noise levels. All of this to say that it's possible to download their data online and do a quick MF analysis (I did that), and with a little bit of work you get a blindingly bright statistical significance of 20-sigma or so. The actual result quoted in the papers was a bit higher. That's a testament to how well the instrument was built and its behaviour understood.