How can machines and humans team up to reveal the hidden mysteries of the Universe – from giant space lasers to distant galaxies? We speak with Alex Andersson about his work to do just that.
Telescopes like JWST or the upcoming Square Kilometre Array have the power to reveal the hidden depths of the Universe. Their images contain not just the stars, galaxies and planets astronomers are looking for, but could contain all manner of hidden gems we have yet to even consider. However these observatories produce so much data it’s impossible to look through it all by hand and even citizen science projects like Zooniverse, which gets hundreds of volunteers to go through the data, could struggle to cope.
Alex Andersson from Oxford University is working on a new machine learning programme that has already revealed a giant space laser (otherwise known as a maser star). We look at how humans and machines can work together to find the undiscovered mysteries of the Universe.
Chris Bramley: Hello and welcome to Radio Astronomy, the podcast from the makers of BBC Sky at Night Magazine. You can subscribe to the print edition of the magazine by visiting www.skyatnightmagazine.com or to our digital edition by visiting iTunes or Google Play.
Ezzy Pearson: Welcome to Radio Astronomy, the podcast from the makers of BBC Sky at Night Magazine. Today we are traveling down to the National Astronomers Meeting being held this year in Cardiff. It’s an annual event where astronomers from around the country gather together to talk about their various different astronomical endeavours.
Hello. I’m at the National Astronomers meeting here in Cardiff today, and today I’m talking to Alex Anderson, who gave a talk about discovering radio transients, using humans and machines. That sounds very exciting. One of the particularly exciting things was that you use a data set from the particularly evocatively named ThunderKAT. So could you tell us what exactly is ThunderKAT and what were you looking for?
Alex Andersson: Yeah, sure. So, ThunderKAT what we do is a survey, taken from the MeerKAT telescope. And ThunderKAT as a name stands for a slightly sort of tortured acronym for the Hunt for Dynamic and Explosive Radio Transients with MeerKAT so really, you know, rolls off the tongue. But I think it was inspired by one of the, one of the PIs of the project. Their husband was a big fan of the Thundercat comic books and so I think it’s kind of been forced in there. But yeah, I think it’s a good name and the whole name of it is to look for different kinds of things that sort of go bump in the night with radio telescopes basically.
So it, we look for all kinds of systems within our galaxy and out of our galaxy, and my work is on searching for interesting transient things that we didn’t necessarily know about beforehand.
Ezzy: So when you say a, a transient thing, what do you actually mean?
Alex: Yeah. It’s a bit, a bit of a catch-all, right?
So just a radio transient in this case it’s just something that we see, you know, when we look in one week with our radio transients, it’s not there. And then, look a few weeks later and it’s suddenly there. So this could be things like nearby stars that sort of flare like our Sun does. This could be distant galaxies that are sort of simulating a little bit and just about anything in between.
So it’s a pretty general method that doesn’t specialise into anything. Just sort of stars, galaxies, anything in between really.
Ezzy: And you searched through your data set using humans and machines. Indeed. Yeah. So the human part of it is a citizen science project. Why are using people particularly good?
Alex: Yeah! So, we have our, our sort of citizen science project called Bursts from Space MeerKAT. Which looks for bursts from space, from MeerKAT. It’s worked really well. And cause our volunteers sort of are able to look through our data in a much more sort of thorough way than say I could do by myself.
Right? So I could say, set some sort of statistical sort of test to say, “oh, if the thing I’m looking for behaves like this or changes like that”, that might be one way of finding it. But sort of previous tests have shown that that often doesn’t get the best results and our volunteers can find things.
That we wouldn’t be able to find with those kind of sort of statistical tests alone. And so they can basically dig through our data, you know, quickly and very efficiently and find lots of interesting things that we wouldn’t have found otherwise.
Ezzy: Mm-hmm. But obviously with… there’s whole new generations of telescopes coming out and one of the big problems I know a lot of astronomers having is just way too much data. So how are you managing to cope with that even when you… even when you’ve democratised it to a thousand people, how are you coping with that?
Alex: Yeah. So this is, this is a really good point, right. And the citizen science work we’ve done so far… it is, it’s very quick over the sort of sky area that we’ve covered with the survey.
But it’s not infinitely scalable, right? It’s, you know, as data rates get more intense and, you know, there’s not a, we can’t expect billions of people looking at my, you know, our data. Even though that would be quite nice. So yeah, so one of the answers is turning to machine learning techniques.
So specifically I’ve been using sort of unsupervised anomaly detection techniques that sort of look to find the anomalous and strange and interesting things in your data. And then I can sort of leverage the, power of our volunteers to see are the things that our sort of machine models find to be interesting and anomalous, are those the same things that our volunteers said were interesting variables and transients?
Ezzy: So you are specifically looking for the things that are weird?
Alex: Yeah, yeah. Sort of. It’s yeah.
Ezzy: The things you’re not expecting to find.
Alex: Exactly. Yeah. So we phrase it as a anomaly detection type thing because you know, in a given patch of sky, maybe 1 or 2% of the things in that region of sky will be the interesting variables and transients.
So the 98% of it is, is haystack and only 1 or 2% is needle. Right. So that’s the kind of thing we’re after. Yeah.
Ezzy: And have you managed to find anything interesting so far?
Alex: Yeah, so, so but the volunteers have done some amazing work. One of my favourite things that they found is this thing called an OH Maser star. So this is… this is not really my area of expertise. So I, they found this and I went, “oh, that’s really cool. Let’s, let’s find out more about that”. And so this is a, star that’s lived through it’s, sort of main middle age and is now sort of in the end of its life and it has a sort of very big, sort of dusty atmosphere around it.And the star is sort of in the middle of that dust and it sort of it basically locates a laser in space by sort of pumping this sort of gas and dust around it with light and yes, it’s a space laser. They basically found.
Ezzy: A giant space laser.
Alex: Yeah, pretty much. And it shows some really interesting behaviour, which we can’t fully explain yet. We’ve got some ideas and we might do some, some more observations to, to figure that out. And that’s been a nice, nice one to, to get involved with our volunteers.
Ezzy: So do you think that one day machines will be able to completely replace humans looking through this kind of data?
Alex: Yeah, that’s a good point. Certainly using machine learning techniques and AI and so on is going to become, you know, ever more important as data gets more and more… But I don’t think it can do everything. There’s, been some work from collaborators of mine and others that have shown that, you know, one of the best things you can do is try and use the machines that things they’re good at. So doing, you know, lots of computation, digging through, you know, vast amounts of data, but then where the humans really excel is where, you know, where things get a bit unexpected or serendipitous, or… Yeah, into the unknown little bit or, and so, you know, there are ways you might be able to prioritise if the machines can do, you know, the heavy lifting and then provide sort of the a smaller sample of the most interesting candidates, , in your data for example.
That’s what the, then the volunteers can do. And there’s been some work that I’ve seen that she shows that the combination of the two provides a sort of unique window that does better than either technique alone, which is I think is really promising. Mm-hmm.
Ezzy: So it’s the, the machine finds the, sort of.. winnows it down and then you have double check it.
Alex: Yeah. I think that’s a really good way of, of doing it. Yeah. Yeah.
Ezzy: And we’ve been talking a lot about the, the data that you’re looking through it. Mm-hmm. But it’s been taken by, by MeerKAT. As you said. Which obviously isn’t a small animal, so could you tell us what exactly is MeerKAT?
Alex: Right. So, MeerKAT is a telescope in South Africa, in the Karoo region of South Africa. And so as in all areas of astronomy, it’s also an acronym. So the KAT part stands for Karoo Array Telescope because it’s an array of telescopes in the Karoo region, and the mayor is, is Afrikaans for more.
So originally the project had seven dishes and now they’ve got more of them. So decided to call it “More KATs” basically. And ,by pure coincidence, I’m sure that’s also happens to be the name of the small loveable animal in the area. But yeah, it’s a fantastic facility that came online only a few years ago and is really, you know, changing the way we can do this kind of science and that’s only gonna get more interesting as it sort of scales up to become part of the SKA in due to course.
Ezzy: Mm-hmm. Which is the Square Kilometre Away, that’s being built across South Africa and Australia. So obviously the work that you are doing is specifically designed for these radio telescopes. Could it be applied to other large data sets as well?
Alex: Yeah, so, so some of my work, particularly the stuff with the machine learning has taken a bit of a leaf out of the book of people who search for optical transients.
So this kind of anomaly selection technique’s used on light curves, how the, you know, brightness of our sources changes over time. That’s something I’ve taken directly from that work. So it is already been used quite a lot in anticipation for things like. The Vera Ruben Observatory and it’s… which is a facility coming online in a few years maybe, or maybe next year. I can’t remember where it’s coming online which is gonna be similarly searching the sky over a long time looking for interesting transient as well. So very much this kind of technique works in, you know, any regime where you’ve got lots and lots of data and want to find interesting, weird things.
Ezzy: Well, it, it certainly sounds fascinating. And because it does have a citizen science aspect to it, is it something that people listening at home could possibly get involved with?
Alex: Oh, yeah. I mean, the more people can get involved, the better. So yeah, our project’s hosted on the Zooniverse website.
So if you just maybe Google something like Zooniverse Burst from Space MeerKAT, you’ll be able to find our project. We’re currently out of data. Our volunteers are actually sort of they’re almost too good in some sense. Like they, they process the data and tell us all about it in more time than we can prepare new data for them.
So we’ve got more data we want to show to them, which we’re hoping to launch next month. So, so keep eyes peeled for that. That’s where it is. Bursts from Space MeerKAT
Ezzy: So if anybody is at home wants to get involved, that’s how you can go about doing that. And thank you very much for talking to us today, Alex.
Alex: Absolutely. Pleasure. Thanks for having me.
Ezzy: Thank you for listening to the Radio Astronomy podcast. Do be sure to subscribe to get all of the latest astronomy news and stargazing tips, and we hope to see you here next time.
Chris Bramley: Thank you for listening to this episode of the Radio Astronomy Podcast from the makers of BBC Sky at Night Magazine. For more of our podcasts, visit our website www.skyatnightmagazine.com or head to iTunes or Spotify.