Speaker
Henry Day-Hall
(desy)
Description
Simulation is a major computational expense in HEP - calorimeter simulation in particular drives the overall energy cost of our physics analysis. Future detectors will include more detailed calorimeters than ever, and the data analysis will require unprecedented statistics from simulation.
Fast generative models developed by our group are redefining the possibilities; creating simulations 100 times more efficient. In this talk I describe key elements of data format and attention masking that facilitate our ultralight ML models. Additionally, I address Jevon's paradox in our field, and consider the importance of realistic metrics to create "good" efficiency.
Author
Henry Day-Hall
(desy)