Go to the Indico Home Page Oldest event Older event Up to category Newer event Newest event Printable version
  • Compact style
  • Indico style
  • Indico style - inline minutes
  • Indico style - numbered
  • Indico style - numbered + minutes
  • Indico Weeks View
    • Switch to the management area of this event
    Login

    EP R&D Software Working Group Meeting

    → Europe/Zurich
    The Zoomiverse

    The Zoomiverse
    Live Notes
    EP R&D Software Working Group Meeting
    Zoom Meeting ID
    62893115044
    Host
    Graeme A Stewart
    Alternative host
    Andre Sailer
    Useful links
    Join via phone
    Zoom URL
    Hide

    Software R&D Working Meeting Minutes

    2021-09-22

    Introduction

    • IRIS-HEP fellowships can be granted to very early career people, even undergraduates. Very useful way to attract talent to our field.
    • ACAT “results” will be known in advance of needing to submit CHEP abstracts (ACAT is 29 November - 3 December).
    • There is also an ML4Sim seminar on Thurdsay 30 September: https://indico.cern.ch/event/1073290/.

    Faster Simulation

    • Different calorimeter materials, but is it a different geometry?
      • Not geometry per-se, but material in this case (slides 8, 12). leading to very different energy deposition profile.
      • But meta-learning model does really use different geometries (from GDML, slide 14)
    • How do the hits/energy deposits come in?
      • Via the Geant4 hits, mapped to the geometry
    • What about “invisible” energy (e.g., using Geant4 tracks)?
      • No, not training with this, only using EM energy deposits at the moment
    • If there’s a new geometry can we try this? New detector concepts.
      • Yes
    • What about multi-dimensional metrics (beyond energy and energy per layer)
      • It’s the next step, but E and E_layer have to first be correct, so that’s where we started
    • How do you represent the different geometries?
      • Using a tokenisation of the GDML features (slide 38)
        • These become the conditioning “hash”
        • But that does not encode any simlarities between any geometries
      • However, this does not encode everything, so we are investigating passing relevant parameters (layers, thickness, material, etc.)
    • Slide 8 geometry, is the particle always incident at (0, 0, 0)?
      • Yes, it is at the moment
      • Open question as to if the training will scale to this being a new input parameter
        • Maybe scaling is better with the ‘learning to learn’ approach?
      • Response from cells can be very sensitive to impact position?
        • Can use a higher granularity approach, with voxelisation finer than cells (although that’s not done right now)
    There are minutes attached to this event. Show them.