Transfer Learning Across Fast- and Full-Simulation Domains in High-Energy Physics

TL;DR

This paper demonstrates that transfer learning from fast to full simulation in high-energy physics improves model performance and reduces training data needs across multiple tasks and architectures.

Contribution

It systematically evaluates transfer learning between simulation domains, showing pretrained models outperform baselines and require less data, promoting reusable scientific assets.

Findings

Pretrained models outperform independently trained baselines.

Transfer learning reduces target-domain training data by about half.

Models generalize across different simulation environments.

Abstract

Machine-learning models in high-energy physics are often trained on simulated data, where fully simulated samples are computationally expensive while fast simulation provides large statistics at reduced realism. In this work, we systematically study transfer learning between fast-simulated and fully simulated datasets in a realistic LHC environment. We consider three representative tasks, signal-background classification, quark-gluon jet tagging, and missing transverse energy reconstruction, using dense neural networks, graph neural networks, and transformer-based architectures. Models are pretrained on ATLAS-like fast simulation and adapted to CMS-like fast simulation and to fully simulated ATLAS Open Data. Across all tasks, pretrained models consistently outperform independently trained baselines and require significantly less target-domain training data, typically reducing the needed statistics by about a factor of two. These results demonstrate that fast simulation can be used to learn robust, reusable representations and motivate publishing trained models as reusable scientific assets beyond large foundation models.