David Miller, University of Chicago - ATLAS, ML

"Black Boxes or Interpretable Models? Applications of Machine Learning, Symmetries, and Domain Knowledge to High-Dimensional Problems in Particle Physics"

Abstract:
The world of artificial intelligence (AI) and machine learning (ML) has undergone what Brian Nord referred to as the "3rd Age of AI” due to the confluence of developments in Algorithms, Computing Resources, and Big Data. Particle Physics has benefitted from, and in many ways strengthened and advanced, progress in AI/ML for decades due to its proliferation of enormous data sets, complex instrumentation, and computing infrastructure. However, there exist both known and unknown deficiencies in our ability to explain “why” some AI/ML models yield a certain result. From my very novice perspective, I will discuss some of the context and common applications of AI/ML in experimental particle physics. I will then focus on three projects ongoing in my group that we believe target important problems relevant to the use of machine learning, symmetries, and domain knowledge in particle physics.