A Lorentz-Equivariant Transformer for All of the LHC

Abstract

We show that the Lorentz-Equivariant Geometric Algebra Transformer (L-GATr) yields state-of-the-art performance for a wide range of machine learning tasks at the Large Hadron Collider. L-GATr represents data in a geometric algebra over space-time and is equivariant under Lorentz transformations. The underlying architecture is a versatile and scalable transformer, which is able to break symmetries if needed. We demonstrate the power of L-GATr for amplitude regression and jet classification, and then benchmark it as the first Lorentz-equivariant generative network. For all three LHC tasks, we find significant improvements over previous architectures.

arXiv comment27 pages, 7 figures, 9 tables. v2: added table 5, improved tagging results. v3: added table 7, incorporate feedback

References in corpus (67)

• The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations
• The anti-k_t jet clustering algorithm
• An Introduction to PYTHIA 8.2
• FastJet user manual
• Gaussian Error Linear Units (GELUs)
• DELPHES 3, A modular framework for fast simulation of a generic collider experiment
• MadGraph 5 : Going Beyond
• Searching for Exotic Particles in High-Energy Physics with Deep Learning
• Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges
• FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
• ParticleNet: Jet Tagging via Particle Clouds
• CaloGAN: Simulating 3D High Energy Particle Showers in Multi-Layer Electromagnetic Calorimeters with Generative Adversarial Networks
• Jet-Images -- Deep Learning Edition
• Deep-learning Top Taggers or The End of QCD?
• Energy Flow Networks: Deep Sets for Particle Jets
• The Machine Learning Landscape of Top Taggers
• Deep-learned Top Tagging with a Lorentz Layer
• Seeing in Color: Jet Superstructure
• Jet-Images: Computer Vision Inspired Techniques for Jet Tagging
• How to GAN LHC Events
• Symbolic Discovery of Optimization Algorithms
• QCD or What?
• Getting High: High Fidelity Simulation of High Granularity Calorimeters with High Speed
• QCD-Aware Recursive Neural Networks for Jet Physics
• An Efficient Lorentz Equivariant Graph Neural Network for Jet Tagging
• Machine Learning and LHC Event Generation
• Performance of top-quark and $W$-boson tagging with ATLAS in Run 2 of the LHC
• On Layer Normalization in the Transformer Architecture
• Flow Matching for Generative Modeling
• Invertible Networks or Partons to Detector and Back Again
• How to GAN away Detector Effects
• Deep Sets
• Event Generators for High-Energy Physics Experiments
• A factorisation-aware Matrix element emulator
• Optimising simulations for diphoton production at hadron colliders using amplitude neural networks
• MadNIS -- Neural Multi-Channel Importance Sampling
• Particle Transformer for Jet Tagging
• Measurements of top-quark pair single- and double-differential cross-sections in the all-hadronic channel in $pp$ collisions at $\sqrt{s}=13~\mbox{TeV}$ using the ATLAS detector
• Two Invertible Networks for the Matrix Element Method
• Event Generation and Statistical Sampling for Physics with Deep Generative Models and a Density Information Buffer
• Modern Machine Learning for LHC Physicists
• Scalars are universal: Equivariant machine learning, structured like classical physics
• Reports of My Demise Are Greatly Exaggerated: $N$-subjettiness Taggers Take On Jet Images
• Towards a Deep Learning Model for Hadronization
• The MadNIS Reloaded
• Explainable Equivariant Neural Networks for Particle Physics: PELICAN
• Stochastic Interpolants: A Unifying Framework for Flows and Diffusions
• Improving Generative Model-based Unfolding with Schrödinger Bridges
• One-loop matrix element emulation with factorisation awareness
• End-To-End Latent Variational Diffusion Models for Inverse Problems in High Energy Physics
• CaloDREAM -- Detector Response Emulation via Attentive flow Matching
• The Landscape of Unfolding with Machine Learning
• Precision-Machine Learning for the Matrix Element Method
• Lorentz-Equivariant Geometric Algebra Transformers for High-Energy Physics
• Differentiable MadNIS-Lite
• Loop Amplitudes from Precision Networks
• Jet Tagging with More-Interaction Particle Transformer
• Towards a data-driven model of hadronization using normalizing flows
• PELICAN: Permutation Equivariant and Lorentz Invariant or Covariant Aggregator Network for Particle Physics
• Clifford Group Equivariant Neural Networks
• Geometric Algebra Transformer
• An unfolding method based on conditional Invertible Neural Networks (cINN) using iterative training
• Optimal Equivariant Architectures from the Symmetries of Matrix-Element Likelihoods
• Describing Hadronization via Histories and Observables for Monte-Carlo Event Reweighting
• Parton Labeling without Matching: Unveiling Emergent Labelling Capabilities in Regression Models
• Fitting a Deep Generative Hadronization Model
• Euclidean, Projective, Conformal: Choosing a Geometric Algebra for Equivariant Transformers

Cited by in corpus (13)

• Accelerating multijet-merged event generation with neural network matrix element surrogates
• Extrapolating Jet Radiation with Autoregressive Transformers
• Amplitude Uncertainties Everywhere All at Once
• BitHEP -- The Limits of Low-Precision ML in HEP
• OmniLearned: A Foundation Model Framework for All Tasks Involving Jet Physics
• Amplitude Surrogates for Multi-Jet Processes
• CaloHadronic: a diffusion model for the generation of hadronic showers
• The Physics Behind ML-based Quark-Gluon Taggers
• CaloClouds3: Ultra-Fast Geometry-Independent Highly-Granular Calorimeter Simulation
• Large Language Models -- the Future of Fundamental Physics?
• TIGER: A Topology-Agnostic, Hierarchical Graph Network for Event Reconstruction
• Stable and Interpretable Jet Physics with IRC-Safe Equivariant Feature Extraction
• jBOT: Semantic Jet Representation Clustering Emerges from Self-Distillation