Joint INFN-UNIMI-UNIMIB Pheno Seminars
Two New Ideas for ML-Theory
by
→
Europe/Zurich
Aula L (Dipartimento di Fisica, Università degli Studi di Milano)
Aula L
Dipartimento di Fisica, Università degli Studi di Milano
Via Giovanni Celoria, 16, 20133 Milano MI, Italia
Description
Machine learning is an exciting new avenue in theoretical particle physics (as I do not have to explain here). I will discuss two recent applications, starting with modern generative networks, which can be used to generate LHC events. Here the key question is how we can control the network and provide uncertainties in addition to the event samples reflecting a learned phase space density. Second, already very simple genetic algorithms allow us to invert the usual tool chain of LHC physics, which starts with often simple formulas and leads to complex numerical tools. I will show how numerically defined functions like optimal observables can be encoded in closed, and correct, formulas using symbolic regression.