| Causal AI in Astrophysics |
| Peter Melchior, Princeton |
| Event Type: Astro Seminar |
|
| Time: 2:00 PM - 3:15 PM |
|
| Location: 726 Broadway, 940, CCPP Seminar |
|
| Abstract: Astronomy and astrophysics are currently undergoing profound changes as the results of two developments: 1) the availability of vast quantities of data from surveys and simulations, and 2) the rapid progress in machine learning and AI. But how should we bridge the gap between data-driven and theoretical descriptions of the Universe? How do we actually learn new aspects of physical systems from data? I will show that introducing causal structure into deep learning architectures creates efficient, robust, useful, and highly interpretable models that respect known physics and reveal unknown phenomena. I will present results from my group on exoplanets, galaxy evolution, and cosmology to demonstrate what we can already achieve and discuss how this approach leads to future astrophysics and AI research. |