HEP for AI

In just the past decade, neural networks have made stunning progress on tasks long thought to be exclusive to humans, but the "hard problem" of artificial intelligence remains: why does a trained neural network give the output it does? In this talk, I will show that an approach to studying neural networks which borrows techniques and perspectives from high-energy physics can make quantitative progress on at least three important facets of this problem: what happens during training, why performance appears to scale predictably and robustly with the amount of training data, and how we can assign error bars to neural network analyses. I will argue that the needs for interpretability and uncertainty quantification in physics applications of machine learning mitigate toward the use of simpler architectures with more predictable performance. I will demonstrate that high-energy physics provides a suite of theoretical tools naturally suited for these bare-bones neural networks, and how the topology and geometry of collider physics data may be used as a testbed for theories of machine learning relevant for data “in the wild”.