Room: Graeme, Leonhard, Alberto, Juan, Joanna, Peter, Florine, Mateusz, Andi, André, Anna
Remote: Swathi, Joshua, Peter Hristov
How long does it take to do the first training?
Speed cf. Geant4?
What happened to GANs?
Andi - DDFastShowerML, re. ATLAS ISF, this was overdesigned and the trigger step had far too many options (topology, for example). Lots of consistency problems. In practice now only geometry is used, and a little bit of a particle triggering. Code complexity is hard to manage. Ask what is it that we really need?
Juan - Are other detectors than ILD supported for the di-photon benchmark. Yes.
Graeme - are diffusion models hitting the mark for physics performance?
Graeme - for training, eventually the "traning from scratch" model would converge? And the training is always on the full reverse diffusion (un-distilled)