THE CHALLENGE

The use of Machine Learning in astronomy is rapidly increasing. All large-scale surveys are addressing the need for automated exploration and reduction of unprecedented amounts of data. The astronomy community has a corresponding need for education and training. However, the scope and dynamism of Machine Learning can make this task seem daunting.

THE SOLUTION

To assist researchers, the Pawsey Supercomputing Centre and ADACS presented Cosmic Machines – a webinar where participants were guided through a hands-on introduction to the use of Machine Learning in observational astronomy. The focus was on gaining an appreciation for the utility of Machine Learning with a short introduction to the fundamental concepts of Deep Learning, and instructions on how to quickly apply it in their own research.

Figure 1: Labelled galaxy images used in training a galaxy classification model.

THE OUTCOME

More than 50 researchers participated and actively engaged in the webinar. After covering the essential elements of Deep Learning, they worked through the steps involved in training a model to label images of galaxies from the Sloan Digital Sky Survey to an accuracy of greater than 90%. Utilizing the fastai framework and Transfer Learning to minimize the time and coding required, the exercise demonstrated clearly that Machine Learning can be an effective and efficient research tool.

 

 

Figure 2: Pre- and post-event survey results showing a marked increase in confidence.

Survey results gathered pre- and post-event display a marked increase in confidence in terms of both conceptual understanding and practical application of Machine Learning (Figure 2), and feedback was overwhelmingly positive:

“It was very interesting, and I got very good introduction about machine learning.” Shintaro Yoshiura, JSPS Overseas Research Fellow.

“I found your talk and the hands-on session very useful. Many thanks for hosting this. We have a student here doing a ML project for gravitational wave research and also praised the webinar highly.” Dr Qi Chu, Research Fellow, OzGrav-UWA.

“I just wanted to say thank you for such a great ML workshop last week. Hope to see more of these workshops in a future. It was short, practical and straight forward. Well done!” Soheil Koushan, PhD Candidate, ICRAR.