Deep Sets for Particle Jets

One of most basic facts about quantum mechanics is that identical particles are indistinguishable.  One of most basic facts about quantum field theory is that only infrared-and-collinear-safe observables can be calculated in a fixed-order expansion.  In this talk, I show how to incorporate both of these facts into a novel machine learning architecture called Energy Flow Networks (EFNs).  EFNs are a special case of a more general architecture called Deep Sets, with the nice feature that one can "open the box" of an EFN to gain insight into what the network has learned.  Using the example of quark/gluon jet tagging at the LHC, I highlight the excellent performance of EFNs and their intuitive visualization.