How astrophysics and machine learning took Michele Sasdelli from Germany to England to NASA to Adelaide

Story written by Dr Sarah Keenihan, AIML

One of the problems with space junk is that it’s really hard to keep track of. 

Yes sure, it’s orbiting around our planet. 

“But objects don’t tend to stay in a fixed position above the Earth,” says Dr Michele (Mike) Sasdelli, Post-Doctoral Researcher at the Australian Institute for Machine Learning (AIML), at the University of Adelaide.

“And even if you can see an object’s trajectory for a brief period, that’s not enough to be able to track it accurately over a long time.” 

Mike is working on a research project where he models how objects move in Earth’s orbit. 

He hopes to be able to better predict the movement of space junk so we can avoid collisions of items in space (there was a near miss in 2020), or anticipate when junk falls back to Earth (like the Chinese space station Tiangong-1 in 2018). 

There are estimated to be thousands of junk items in space, and over a million if you include the pieces smaller than 1cm in size. 

Looking for planets 

This project is not Mike’s first in the realm of space. 

Prior to commencing at AIML, he spent three months working at NASA with other machine learning and astronomy experts. He was engaged through the Frontier Development Lab, which applies AI technologies to scientific research and solving human problems. 

“We were developing algorithms to detect exoplanets,” Mike said. “The data we used came from the TESS telescope, which is so sensitive it can detect subtle changes in the brightness of stars due to a planet passing in front them.”  

TESS is short for Transiting Exoplanet Survey Satellite. Exoplanets are planets that orbit stars outside of our solar system, and some could theoretically support life. 

“We published a couple of papers based on our work,” said Mike. “And now other scientists are building up these ideas further.” 

Frontier Development Lab has published a short video that summarises this project. 

Theory of learning

After growing up in Italy, Mike completed an astrophysics PhD in Germany. 

“But since then I’ve moved more into machine learning, and spent less time in physics,” he said.  

After graduating, Mike spent several years working in UK computer vision company Cortexica, which is now part of Zebra Technologies. 

He then came to AIML in Adelaide, with his stint at NASA on the way. 

The majority of Mike’s current research activities focus on deep learning, and theory of learning. 

Deep learning is a kind of artificial intelligence (AI) that mimics some of the features of the human brain to detect objects, recognize speech, translate languages and make decisions. It is able to learn from different kinds of data, and without human supervision. 

“Right now we have complicated AI systems that are able to learn quite detailed tasks, but actually we don’t always know exactly how they manage to do this,” Mike says. 

“I want to have a greater understanding of how deep learning actually works.” 

“Once we understand is better, we can make improvements in deep learning systems to improve computer vision, ‘self-driving’ car technology, language recognition and more,” adds Mike.