Shared Data and Algorithms for Deep Learning in Fundamental Physics

Abstract

We introduce a Python package that provides simple and unified access to a collection of datasets from fundamental physics research—including particle physics, astroparticle physics, and hadron- and nuclear physics—for supervised machine learning studies. The datasets contain hadronic top quarks, cosmic-ray-induced air showers, phase transitions in hadronic matter, and generator-level histories. While public datasets from multiple fundamental physics disciplines already exist, the common interface and provided reference models simplify future work on cross-disciplinary machine learning and transfer learning in fundamental physics. We discuss the design and structure and line out how additional datasets can be submitted for inclusion. As showcase application, we present a simple yet flexible graph-based neural network architecture that can easily be applied to a wide range of supervised learning tasks. We show that our approach reaches performance close to dedicated methods on all datasets. To simplify adaptation for various problems, we provide easy-to-follow instructions on how graph-based representations of data structures, relevant for fundamental physics, can be constructed and provide code implementations for several of them. Implementations are also provided for our proposed method and all reference algorithms.

Machine learning and LHC event generation

“…Publication of ML models for their reuse is not yet standard in the particle physics community. Examples where trained networks have been published in ONNX format for future reuse are the DNNLikelihood [123], a package for cross-disciplinary training of discriminator networks [124], and the ATLAS search for R-parity-violating supersymmetry [125,126], the latter also being available in the ATLAS SimpleAnalysis framework. However, detailed documentation for instance of the input variables is missing.…”

Application of graph networks to background rejection in Imaging Air Cherenkov Telescopes

“…In this work, we investigate two different formulations that use the spatial ansatz, which is very flexible and has already been successfully utilized in physics [19][20][21]29].…”

A Normalized Autoencoder for LHC Triggers

“…The first test will be top vs QCD, discussed in Sec. 3, where we use the top-tagging dataset [62][63][64], also used for the AE in Ref.…”

“…In Sec. 3 we apply the NAE to the top tagging dataset [62][63][64] and show that, for the first time, the NAE identifies anomalous top jets and anomalous QCD jets symmetrically and with high efficiency. Next, we target two challenging dark jet signals [15] and confirm the excellent performance of the NAE and its relative independence of the jet image preprocessing in Sec.…”