NEXUS: Lightweight Foundation Model for Collider Physics and Beyond

Liangyu Wu, Qibin Liu, Alexander Yue, Julia Gonski· July 31, 2026 View original

Key takeaways

  • NEXUS is a lightweight foundation model for collider physics.
  • Unsupervised pre-training improves accuracy with small labeled datasets.
  • It demonstrates successful multi-domain adaptation to other sciences.
  • The model enables power-efficient, real-time AI applications.

Who benefits

Scientific ResearchEnergyEnvironmental MonitoringHealthcareAerospace

Summary

Researchers introduce NEXUS, a lightweight foundation model (3M parameters) for collider physics that leverages unsupervised pre-training on particle track features. It achieves improved accuracy on downstream tasks with small labeled datasets and demonstrates multi-domain adaptation to gravitational waves, flood forecasting, and neural activity.

This paper presents NEXUS, a lightweight foundation model specifically designed for collider physics, yet capable of multi-domain adaptation. With approximately 3 million parameters, NEXUS employs a fully connected autoencoder architecture and undergoes unsupervised pre-training on a vast dataset of charged particle track features from the Large Hadron Collider. This pre-training enables the model to achieve improved accuracy on downstream collider analyses, such as kinematic regression and event classification, even when only small labeled datasets are available, outperforming equivalent architectures trained from scratch. Beyond its primary application, NEXUS demonstrates significant benefits in latent space interpretation and successful adaptation to entirely different scientific domains. The research highlights its applicability to tasks in gravitational wave analysis, flood forecasting, and neural activity prediction. Crucially, NEXUS offers computational simplicity compared to larger transformer models, paving the way for power-efficient inference and real-time or edge applications of foundation models in diverse scientific experiments.

Why it matters

Developing lightweight, adaptable foundation models is crucial for deploying AI in resource-constrained environments and for accelerating scientific discovery across multiple disciplines without requiring massive computational resources.

How to implement this in your domain

  1. 1Evaluate NEXUS or similar lightweight foundation models for specific scientific or industrial applications requiring efficient inference.
  2. 2Explore unsupervised pre-training strategies on large unlabeled datasets relevant to your domain to build domain-specific foundation models.
  3. 3Investigate the potential of adapting pre-trained models from one scientific domain to another within your organization.
  4. 4Develop edge computing solutions leveraging lightweight foundation models for real-time data analysis in remote or power-limited settings.
  5. 5Train data scientists and researchers on the principles of foundation model adaptation and efficient model deployment.

Original post by Liangyu Wu, Qibin Liu, Alexander Yue, Julia Gonski

"arXiv:2607.27501v1 Announce Type: new Abstract: We present a lightweight approach to foundation modeling (\textbf{NEXUS}) that leverages pre-trained learning from collider physics data towards out-of-domain tasks in other scientific datasets, using a fully connected autoencoder m…"

View on X

Originally posted by Liangyu Wu, Qibin Liu, Alexander Yue, Julia Gonski on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

New Framework Improves Partial Multi-View Clustering Performance.

DAS-PMVC is a novel framework for partial multi-view clustering that addresses view asymmetry and irrelevant samples by leveraging dual alignment and structure enhancement. It uses anchor graph structure alignment, structure-enhanced feature learning, and a dual alignment strategy to achieve superior clustering performance on various datasets.

Shubin Ma, Liang Zhao, Chuanye He, Zhenjiao Liu, Liang Zou, Lin Yuanbo Wu, Yu ShaoJul 31, 2026
AI Engineering & DevToolsAI Research

Dual Teachers Improve Adversarial Robustness and Accuracy.

This work extends Information Bottleneck Distillation (IBD) by introducing a "clean teacher" alongside a robust teacher to improve the robustness/accuracy tradeoff against adversarial attacks. The proposed method transfers features from both teachers to a student model, achieving better clean accuracy while maintaining adversarial robustness, outperforming original IBD and competing with state-of-the-art approaches.

Vincent Ryusuke Takahashi, Yoshinari Takeishi, Jun'ichi Takeuchi, Kave SalamatianJul 31, 2026
AI Engineering & DevToolsAI Research

Dynamic Batch Sizes Improve Large Language Model Training Efficiency.

This paper proposes a new approach to deep learning dynamics, deriving joint scaling laws for loss based on both learning rate and batch size schedules. It introduces an optimal dynamic batch size schedule that consistently outperforms static batch size baselines, highlighting its importance for large language model training.

Jiaxiang Li, Zhiqi Bu, Shiyun XuJul 31, 2026