AI finds hidden anomalies in the Hubble archive

19:57This was This was a bad one, guys. >> This looked like Ferrari at F1 two weeks ago. Not not a good start to the season. >> looked like Aston Martin. What's the other one? Haas? Is Haas doing anything? >> Haas is doing okay. They They've got a Ferrari engine. They're doing okay. Yeah, they're doing okay. >> Okay, so it is simply it is simply the Ferrari. This is something again we're trying to see if you all enjoy. Let us know in the comments how many you got? Are you smarter than scientists? Probably on this one. And we'll bring you back we'll bring it back for next week's rundown. If you want it more, you got to let us know you want it more. But in the meantime, before our next episode game

20:37show, we are going to go to our third story which is about hidden cosmic anomalies in Hubble's archive discovered by AI. Researchers at the European Space Agency, ESA, published in Astronomy and Astrophysics a new AI system called Anomaly Match. Very on the nose. That scanned approximately 99.6 million images cutouts from the entire Legacy Hubble archive in two to three days, which is an unbelievably short period of time. And what did we find? Yeah. So, for background, machine learning and AI

21:19systems are very good at something called anomaly detection, right? Which is something that is out of the ordinary. Just because of the way that they learn data and they learn the data space, there's ways that they can group similar items together in their latent space and they can figure out outliers. It's kind of a I mean, if you think about like rudimentary clustering algorithms where, you know, you project your data into some kind of subspace and then in this case, let's say, "Oh, normal spiral galaxies are over here. Normal barred galaxies are over here." There's going to be little tiny sectors where you find extremely rare objects. And machine learning algorithms can do this

21:59very, very efficiently. Now, what these scientists did from the ESA was apply that method to the Hubble legacy archive, which is over the past 35 years, the Hubble Space Telescope has just been taking image after image, right? And one of the great things about the Hubble Space Telescope is all of their images are now public access. Okay? Um if you're, you know, taking something today, then perhaps it's not public access. You've got like 6 months or something to like grind out as many papers as you can, but pretty soon it's going to go into the public archive. Mhm. So, all of these images were cut out into 100 million

22:40different little image cutouts that are measuring just a few dozen pixels on a side.