Cosmology seminar [extra]: Interpretable Machine Learning for Physics

Would Kepler have discovered his laws if machine learning had been around in 1609? Or would he have been satisfied with the accuracy of some black box regression model, leaving Newton without the inspiration to discover the law of gravitation? In this talk I will discuss problems with the use of industry-oriented machine learning algorithms being used in the natural sciences. I will describe recent approaches I have developed with collaborators for building interpretable machine learning algorithms for science, largely based on a mix of symbolic learning and neural networks. I will discuss the inner workings of my open-source symbolic regression library, PySR (astroautomata.com/PySR/), which forms a central part of this interpretable learning toolkit, and give an interactive tutorial. Finally, I will present examples of how these methods have been used in the past two years in scientific discovery, and outline some current efforts.