Around the world various teams of astronomers and industry experts are looking into how machine learning can be applied to advanced processing techniques for the SKA in the future, tapping into the fast-growing international market around artificial intelligence (AI) and its potential applications.
The Alan Turing Institute is the UK’s national institute for data science and artificial intelligence. Thirteen leading UK universities are members of the institute, which is headquartered at the British Library in London. The AI Fellowship is funded by the UK government’s Department for Business, Energy and Industrial Strategy.
Radio astronomer Prof. Anna Scaife, head of the Jodrell Bank Interferometry Centre of Excellence at the University of Manchester in the UK, is one of those exploring how machine learning can help to tackle the enormous challenge of analysing SKA data.
In October 2019, Anna was awarded an Alan Turing Institute Artificial Intelligence Fellowship, which supports ambitious and transformative research in the field. Only a few months earlier, she received the UK Royal Astronomical Society’s Jackson-Gwilt Medal, awarded for advances in astronomical instrumentation and techniques. Anna also runs training programmes supporting students from southern Africa and Latin America to pursue graduate degrees in the UK focusing on big data and data intensive science.
We caught up with her to hear how the fellowship will assist her work on the SKA, and why machine learning will be essential for SKA science.
Congratulations on being awarded an Alan Turing Institute AI Fellowship – what will this mean for you and your work?
Thanks very much! I’ve been working with machine learning in astronomy for a while now and although it’s always been seen as potentially very useful for big data facilities like the SKA there are many aspects of standard machine learning that just aren’t yet suited to the kind of robust scientific analysis that SKA science will require.
This can be because we don’t have the right data to train the algorithms available yet, or it can be because the algorithms don’t deliver all of the information we need to make statistically rigorous analyses – or that the computational cost of running the algorithms on the volumes of data the SKA will produce is so large that it’s simply not possible to use them effectively.
This fellowship will allow me to start tackling these issues in detail, in partnership with IBM Research. I’ll be looking at how existing AI techniques can be adapted and extended for SKA science, and also to start developing new AI approaches inspired by the needs of SKA data challenges.
The fellowship will mean you’ll spend part of your week at SKA HQ – what will that involve?
A lot of the work that I’ll be doing at the HQ will involve interacting with the science team and the team developing the network of regional centres. Even seemingly routine tasks like identifying, classifying and cataloguing astronomical objects from SKA images will need machine learning solutions when you’re expecting to detect tens of millions of astrophysical systems! And making sure that those classifications are unbiased when they’re going to be used for statistical analyses will require carefully constructed and tested AI.
I also intend to work with the computing and software team for the telescopes, because although the majority of the AI applications that I’ll be working on will be for post-processing, there are a number of areas within the design of the telescope itself that could potentially benefit from assistive machine learning, such as monitoring the SKA’s components to spot potential faults before they happen, or perhaps maintenance scheduling.
There are also some components of the front-end computing where AI is expected to be an integral part of the data handling, namely the pulsar search within the SKA’s Central Signal Processor and Science Data Processor. Being based at the HQ for a few days a week will enable me to interact with teams working across the whole system on a regular basis, which will be great.
“Radio astronomy has always hovered on the boundary between science and engineering, and the SKA exemplifies the interdisciplinary nature of that relationship.”
You’ve been involved in the SKA for many years – which areas have you been working on?
Most of my work for the SKA has been around the standard radio astronomy processing and how we can make it work at SKA-scales, specifically looking at the calibration and imaging algorithms. More recently I’ve been helping to build a design for the European SKA regional centre, where the scientific post-processing for the telescope will happen for SKA scientists based in Europe.
This work required me to look in detail at what that post-processing would involve and how we can extract scientific knowledge from SKA data. A lot of what we need to do for that science will require machine learning and that will happen (mainly) at the regional centres, close to the global scientific community, rather than at the telescope itself.
Artificial Intelligence vs Machine Learning
Machine learning is a subset of AI, and refers to the process of making machines learn from experience, and alter their algorithms accordingly in order to create better, more accurate results. As such, it is an enabler of Artificial Intelligence, the broader effort to get computers to act “intelligently”.
Recently, researchers in Italy, Germany, Australia, China and Iran employed machine learning to identify astronomical sources in the SKA’s first Science Data Challenge, one of several areas where it will likely be invaluable.
In your opinion, what’s so special about the SKA project?
The collaborative aspect of the SKA project is very special. Radio astronomy has always hovered on the boundary between science and engineering, and the SKA exemplifies the interdisciplinary nature of that relationship.
It’s great to work in a project that brings people together, not only from different disciplines but also from all over the world. The project allows you to work and interact with scientists and engineers from a wide range of countries and institutes, with an even wider array of nationalities.
You’re now a professor of radio astronomy but how did it all begin – how did you get into astronomy?
By accident. I spent most of my childhood wanting to be an archaeologist, but somehow I ended up as a radio astronomer. I was never particularly excited by more traditional optical astronomy. Stargazing always seemed like a somewhat cold and uncomfortable way to spend the evening!
However, the first time I had the opportunity to work with a radio telescope, I was hooked. The idea that the sky is full of structures that are invisible to our eyes, but that can be mapped using radio waves is amazing to me. I find the radio Universe to be far more exciting than the visible Universe!
In 2014, you were selected by the World Economic Forum (WEF) as one of thirty scientists under the age of 40 for your contribution to advancing the frontiers of science, engineering or technology in areas of high societal impact. That is very impressive. Tell us more!
It was pretty exciting. The WEF supports an international community of Young Scientists, who are selected annually. I found it fascinating to meet these other scientists from across a broad range of disciplines, but perhaps even more so to have the opportunity to participate in debates with leaders in other areas – politics, business and philanthropic.
By throwing all of these people together, the WEF gives you a great perspective on your own role in the world. You come away with a renewed sense of purpose and a head full of ideas. I have certainly learned more from the WEF than they have from me, I think!
“There are very few projects in science generally, not just in astrophysics, that have data sizes like the SKA.”
“Artificial Intelligence will be a key component to aid interpretation of the large datasets that SKA will deliver to its science community, so I’m delighted that Anna will be able to spend time researching the use of machine-learning techniques in the radio astronomy domain.
We need to understand how AI can help us, but also to see where it needs to be improved to prevent biases from creeping into results, so having expert scientifically rigorous researchers such as Anna working in this field is very exciting.”
Dr. Rosie Bolton, SKA Regional Centre Project Scientist
We often say the SKA is a Big Data project, but what potential applications are there beyond astronomy, which might affect people’s everyday lives?
These days the lives of most people in the developed world are driven by data – even if they don’t realise it. We are constantly receiving and emitting data through a variety of mechanisms, be they social, commercial, medical, financial or otherwise. The art of creating actionable outcomes from these data has so violently changed the way we live our lives that it’s now known as the Big Data revolution. And this is a revolution that has only just begun – the volume, velocity and variety of the data now available to us are almost so overwhelming that the mechanisms of dealing with it need to constantly evolve.
The SKA is a project that drives that evolution. There are very few projects in science generally, not just in astrophysics, that have data sizes like the SKA. This means that, to even make it work, we need to change the way we analyse, transport and store those data. And nearly every innovation that we make for the SKA can be translated to data elsewhere.
The data innovations from the SKA are likely to change the way that data is used across a wide range of applications, but most people will probably never know about it – in the same way that nearly a billion people use Wi-Fi every day without realising that it was also born out of radio astronomy.
See Anna in the 2016 SKA trailer:
Also in this section