Resources for The first MLQC4FC Training School: Tutorial on Generative Models for High-Energy Physics

Abstract

This record contains the resources accompanying the hands-on tutorial "Diffusion Models as High-Energy Physics Surrogates", prepared for The first MLQC4FC Training School. The files are intended for educational purposes and are used in the accompanying Jupyter notebook and GitHub repository. Contents dataset_2_1_10k.hdf5 A reduced version of the Fast Calorimeter Simulation Challenge dataset 2. Contains 10,000 calorimeter showers together with their corresponding incident particle energies. The dataset has been reduced from the original 100,000 showers to make it more suitable for use during the hands-on tutorial, where participants need to download and process the data within a limited amount of time. The file retains the same data structure as the original dataset, with the following datasets: incident_energies: shape (10000, 1) showers: shape (10000, 6480) This reduced dataset is provided specifically for the tutorial and is not intended to replace the original dataset. Checkpoint files Pre-trained model checkpoints used during the tutorial to demonstrate checkpoint loading, resume training, and sample generation without requiring participants to train the model from scratch. These resources are designed for the accompanying hands-on tutorial and are not intended as a benchmark dataset. For the tutorial notebook and source code, please refer to the associated GitHub repository. Usage The reduced dataset is used by the accompanying Jupyter notebook to demonstrate the training and application of diffusion models as surrogate models for high-energy physics simulations. For the tutorial, participants should use dataset_2_1_10k.hdf5 to avoid the large download and computational requirements of the full dataset. The original dataset can be found in the Fast Calorimeter Simulation Challenge resources and should be used for full-scale studies and benchmarks.