New self-supervised pre-training method for LHC foundation models unveiled

Researchers have developed a novel self-supervised pre-training method for foundation models at the Large Hadron Collider (LHC). This data-driven approach utilizes the energy mover's distance (EMD) to pair events based on their similarity, eliminating the need for traditional data augmentations. By matching distinct events by similarity, the method learns invariance without altering event fidelity or requiring computationally intensive simulations. Experiments on QCD jets demonstrate that this augmentation-free pairing method can produce semantic jet embeddings with downstream discrimination power comparable to or exceeding existing augmentation-based baselines. AI

IMPACT This method could improve the efficiency and effectiveness of training foundation models for high-energy physics research.

RANK_REASON The cluster contains a research paper detailing a new method for self-supervised pre-training. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• CatalyzeX Code Finder for Papers
• CORE Recommender
• DagsHub
• Energy Mover's Distance
• Gotit.pub
• Hugging Face
• Influence Flower
• Large Hadron Collider
• QCD jets
• ScienceCast
• Self-distillation bridges distribution gap in language model fine-tuning

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →