Machine Learning in Astronomy and Physics

The Data Exchange Podcast: Dr. Viviana Acquaviva on the impact of machine learning and data science on her research and teaching.

Subscribe: AppleAndroidSpotifyStitcherGoogleAntennaPodRSS.

This week’s guest is Dr. Viviana Acquaviva, Professor in the Physics Department at the CUNY NYC College of Technology and at the CUNY Graduate Center. She is an Astrophysicist with a strong interest in Data Science and Machine Learning.

Download the 2021 NLP Survey Report and learn how companies are using and implementing natural language technologies.

Viviana is currently writing a book for Princeton University Press entitled “Machine Learning techniques for Physics and Astronomy”. This conversation is focused on applications of machine learning and data science to physics, their impact on her research, and how the rise of ML has impacted her teaching and research.

Viviana Acquaviva:

Before machine learning we had to do everything with traditional inference, which basically means that every time we do forward modeling, like every time, we have to come up with some sort of model of observable quantities. We needed to go from the initial condition, like the parameters we need to measure all the way to observable, to what we wanted to measure.

… We needed to be able to simulate how the observed galaxy would be at different ages. This is the thing that is very difficult because the objects that we work with, like galaxies are extremely complex. We needed to write down, or to come up with a picture of a spectrum of a galaxy of a given age. You basically take a very complex system, and you simplify it to what physicists think is important. This comes with all sorts of biases, and then you still have to conduct simulations for many values of the age of the galaxy, and then you will pick the one that looks most like data.

… Most importantly, you know, you pick what variables matter. I think this is the really nice thing that machine learning gave us – a tool to learn representations from data. This is one of the parts that I’m most excited about now.  We can work with data at the image level, we can just sort of like study them, and try to see if we can learn representations that are useful.

Related content:

Subscribe to our Newsletter:
We also publish a popular newsletter where we share highlights from recent episodes, trends in AI / machine learning / data, and a collection of recommendations.

[Image: star_trails_night_long_exposure_starlight_space_rotation_astronomy from pxhere.]