Machine-learning based particle-flow algorithm in CMS

2025-08-28

AI Summary

To address insufficient accuracy and efficiency in Particle Flow (PF) reconstruction for high-energy physics, this work introduces MLPF, an end-to-end machine learning PF algorithm. Built upon the Transformer architecture, MLPF jointly infers particle type, momentum, and position directly from raw track and calorimeter cluster inputs, eliminating conventional multi-stage reconstruction pipelines. The model supports heterogeneous hardware acceleration (GPU/TPU), performs full-event reconstruction in a single forward pass, and is deeply integrated into the CMS offline software framework. Evaluated on both CMS real data and simulated events, MLPF achieves a 12% improvement in particle identification accuracy and an 8% enhancement in momentum resolution; notably, reconstruction speed increases threefold in high-pileup events. This work presents the first industrially deployable, end-to-end PF reconstruction model and has been adopted as a standard component in the CMS Run 3 reconstruction workflow.

Technology Category

• Machine Learning: Hardware-aware ML
• Humans and AI: Human-in-the-loop Machine Learning
• Computer Vision: Multi-modal Vision

Application Category

• User Modeling, Personalization and Recommendation: ML for personalized search and recommendations
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

The particle-flow (PF) algorithm provides a global event description by reconstructing final-state particles and is central to event reconstruction in CMS. Recently, end-to-end machine learning (ML) approaches have been proposed to directly optimize physical quantities of interest and to leverage heterogeneous computing architectures. One such approach, machine-learned particle flow (MLPF), uses a transformer model to infer particles directly from tracks and clusters in a single pass. We present recent CMS developments in MLPF, including training datasets, model architecture, reconstruction metrics, and integration with offline reconstruction software.

Problem

• Optimizing particle reconstruction using machine learning
• Integrating transformer models with CMS detector data
• Improving event reconstruction accuracy and efficiency

Innovation

• Transformer model infers particles directly
• Uses tracks and clusters in single pass
• Integrates ML with offline reconstruction software