Redirecting to original paper in 30 seconds...

Click below to go immediately or wait for automatic redirect

arxiv_ml 85% Match Research Paper High-energy physicists,Particle physicists,Geometric deep learning researchers,Machine learning engineers in scientific domains 1 week ago

Lorentz Local Canonicalization: How to Make Any Network Lorentz-Equivariant

computer-vision › 3d-vision
📄 Abstract

Abstract: Lorentz-equivariant neural networks are becoming the leading architectures for high-energy physics. Current implementations rely on specialized layers, limiting architectural choices. We introduce Lorentz Local Canonicalization (LLoCa), a general framework that renders any backbone network exactly Lorentz-equivariant. Using equivariantly predicted local reference frames, we construct LLoCa-transformers and graph networks. We adapt a recent approach for geometric message passing to the non-compact Lorentz group, allowing propagation of space-time tensorial features. Data augmentation emerges from LLoCa as a special choice of reference frame. Our models achieve competitive and state-of-the-art accuracy on relevant particle physics tasks, while being $4\times$ faster and using $10\times$ fewer FLOPs.
Authors (7)
Jonas Spinner
Luigi Favaro
Peter Lippmann
Sebastian Pitz
Gerrit Gerhartz
Tilman Plehn
+1 more
Submitted
May 26, 2025
arXiv Category
stat.ML
arXiv PDF

Key Contributions

Introduces Lorentz Local Canonicalization (LLoCa), a general framework to make any neural network Lorentz-equivariant, overcoming the limitation of specialized layers. It enables LLoCa-transformers and graph networks, allowing propagation of space-time tensorial features and naturally incorporating data augmentation. Models achieve competitive accuracy while being significantly faster and more efficient.

Business Value

Accelerates scientific discovery in high-energy physics by enabling faster and more efficient analysis of experimental data, potentially leading to breakthroughs and reduced computational costs for research.

Paper Metadata

Innovation Type

Algorithmic Framework

Deployment Feasibility

Moderate. Requires expertise in geometric deep learning and physics applications. Integration into existing HEP analysis pipelines is feasible.

Limitations Addressed

The reliance on specialized, architecture-specific layers for Lorentz-equivariance in neural networks, limiting flexibility and efficiency.

Performance Gains

Models are $4\times$ faster and use $10\times$ fewer FLOPs compared to previous approaches, while achieving competitive/SOTA accuracy.

Technical Tags

Lorentz EquivarianceNeural NetworksHigh-Energy Physics (HEP)Particle PhysicsGeometric Deep LearningLocal CanonicalizationTransformersGraph NetworksSpace-Time TensorsData Augmentation

Research Topics

High-Energy PhysicsGeometric Deep LearningParticle PhysicsMachine LearningPhysics-Informed AI

Methods & Architectures

Lorentz Local Canonicalization (LLoCa)Equivariant reference framesLLoCa-transformersLLoCa-graph networksGeometric message passingTensorial feature propagation LLoCa-transformersLLoCa-graph networksGeneral backbone networks

Applications & Tasks

High-Energy Physics Particle Physics Scientific Machine Learning 3D Geometric Analysis Making arbitrary networks Lorentz-equivariantDeveloping general frameworks for Lorentz-equivariant NNsEfficiently processing space-time tensorial data Particle identificationEvent classificationPhysics simulation analysisGeometric deep learning on space-time data

Datasets & Benchmarks

Benchmarks
Competitive and state-of-the-art accuracy on relevant particle physics tasks
AccuracySpeed (FLOPs)Efficiency (FLOPs)

Related Fields

High-Energy PhysicsParticle PhysicsGeometric Deep LearningMachine LearningTheoretical Physics

Keywords

Lorentz equivarianceNeural networksHigh-energy physicsParticle physicsGeometric deep learningLLoCaTransformersGraph networksSpace-time tensorsData augmentationScientific machine learning

Academic Context

#High-Energy Physics#Geometric Deep Learning#Particle Physics#Machine Learning#Physics-Informed AI

Technology Stack

Frameworks & Libraries
PyTorchTensorFlow
Programming Languages
Python

Commercial Potential

Potential Products
General-purpose Lorentz-equivariant deep learning librariesAccelerated analysis tools for HEP experiments
Target Industries
Scientific Research (Physics)Technology (AI for Science)
Use Case Examples
Analyzing collision data from particle accelerators (e.g., LHC)Classifying particle typesSimulating particle interactions
Competitive Edge
Offers a general, flexible, and efficient framework for achieving Lorentz-equivariance, surpassing previous specialized layer approaches.
Market Opportunity
Niche but critical within the high-energy physics research community.
Revenue Models
Licensing to research institutionsintegration into scientific software platforms.

Resource Requirements

Compute Needs
High (for training deep learning models on HEP data)
Data Requirements
Particle physics datasets (simulated or experimental).
Deployment Constraints
Requires specialized knowledge for application in physics analysis pipelines.
Scalability
The framework is designed to be general and applicable to various backbone networks, suggesting good scalability.

Production Readiness

Maturity Level

Research/Development

Time to Market

2-4 years

Patent Potential

Moderate

View Full Paper Back to Papers