Choose timezone
Your profile timezone:
Powered by Indico
v3.3.5
Tristan presented ideas to speed up the CTA simulation. By changing the input spectrum one could avoid regions of too low statistics. An early termination of the simulation may not work well for CTA because there is a high chance of seeing a signal in at least one of the telescopes for any primary particle.
Jakob showed approaches to speed up simulations at IceCube. Speed ups of a factor ~10 can be reached by aborting simulation with a probability given by the energy fraction of a muon with respect to the primary particle.
Boyang presented a graph neural network to predict if a generated events passes a selection after detector simulation and reconstruction at Belle II. A speedup factor of ~2 is achieved if the classifier output is used as a probability to keep an event.
It was discussed what the problems and solutions have in common and where they differ. In all cases a definition of importance is needed which is given by the physics case. In case of CTA and IceCube the importance is a continuous function, for Belle II it is binary. For Belle II the input features are a tree structure, for CTA and IceCube the input is so far one-dimensional. For CTA and IceCube a partial stopping of the simulation may be a viable approach. As the atmosphere changes constantly the conditions for CTA are not stable and one cannot use models trained on large amounts of old data.
Thomas will request a mailing list for KISS-B2 and send around a poll for a meeting in about 2 to 3 months.