FLC-QU-hep/PointCountFM-multi-geometry
Updated
AI & ML interests
Fast simulation of calorimeter showers with generative models
Recent Activity
View all activity
Papers
Transferable Fast Calorimeter Shower Generation via Multi-Geometry Pre-training
CaloClouds3: Ultra-Fast Geometry-Independent Highly-Granular Calorimeter Simulation
Organization Card
FLC-QU-hep
Machine learning for particle physics at Universität Hamburg and DESY. We develop generative models for fast simulation of calorimeter showers and study how they transfer between detector geometries. Code is on GitHub.
Multi-geometry pre-training
A point cloud shower model pre-trained on a pool of calorimeter geometries, then fine-tuned to a new detector with a fraction of the data and compute of training from scratch (arXiv:2608.18233).
| Repository | Content |
|---|---|
AllShowers-multi-geometry |
Pre-trained shower models, SimpleBox and LEMURS pools |
PointCountFM-multi-geometry |
Pre-trained per-layer point-count models, same two pools |
calorimeter-showers-multi-geometry |
Dataset card for the training and test data |
Data (465 GB) at doi:10.25592/uhhfdm.19103,
code on the multi-geometry branch of
AllShowers and
PointCountFM.
More
Full list of papers, code and datasets on the GitHub organization page.