EVENET: Event-Level Foundation Model for Collider Data Analysis

Nature Machine Intelligence

Overview

EVENET is a foundation model pretrained on 500 million simulated collision events to provide a reusable backbone for diverse particle physics analyses. It outperforms state-of-the-art baselines in low-statistics regimes and demonstrates successful transfer to real experimental data. This approach reduces the need for task-specific model optimization in high-dimensional collider experiments.

In Depth

Key Findings

EVENET demonstrates superior performance over established and state-of-the-art baselines in searches for exotic Higgs decays and heavy resonances, showing particular strength in low-statistics regimes. The model’s utility was validated through the rediscovery of the ϒ meson using CMS Open Data, confirming its ability to transfer from simulation to real experimental data. Additionally, the model enables the precise extraction of quantum-correlation observables that remain stable under tested systematic variations.

Technical Details

The model is designed as an event-level foundation model that treats each collision as an unordered particle cloud. It utilizes a shared point–edge transformer backbone to process variable-length sets of physics objects. Pretraining was conducted on 500 million simulated events using a two-stage strategy that combines self-supervised learning (SSL) with physics-informed supervision.

EVENET supports multiple interchangeable task heads, including:

• Process classification
• Object-to-resonance assignment
• Final-state segmentation
• Diffusion-based reconstruction of undetected or masked particles

Background & Context

High-energy physics, including experiments at the Large Hadron Collider (LHC), Belle II, and DUNE, generates massive volumes of high-dimensional data. Traditionally, researchers have relied on task-specific optimization, designing and training separate models for every new analysis. This paradigm creates significant bottlenecks in expert effort and computing resources, a challenge expected to intensify with the High-Luminosity LHC (HL-LHC). While previous foundation model attempts focused on individual jets or narrow physics processes, EVENET expands this scope to full collision events.

Significance & Outlook

By shifting from task-specific models to a “pretrain-then-transfer” paradigm, EVENET establishes a reusable backbone for collider analyses. This reduces the requirement for extensive labeled data for every new task and minimizes the turnaround time for downstream adaptation. The investment in a substantial one-time pretraining phase allows for more efficient analysis across diverse observational regimes, including fully and partially reconstructed data.

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.

• Your email and selected fields are used only to deliver the newsletter.
• We never share your information with third parties.
• You can unsubscribe anytime via the link in each email.

See our Privacy Policy for details.

Agree & Continue ▶

Takes about a minute · Unsubscribe anytime