Skip to main content

Research timeline

Related research and updates

Public articles linked to the same research event.

arXiv

Researchers recast particle track fitting as a language translation task, and a decoder-only transformer improved track-parameter accuracy by an order of magnitude on synthetic data while identifying multi-track events with 98% accuracy

This exploratory study formulates particle track fitting as translating the language of detectors into the language of physics, designs a simple decoder-only transformer with a custom tokenizer, and reports that on synthetic datasets it outperformed a simple regression technique by an order of magnitude in track-parameter accuracy while being on par with RANSAC, identified multi-track events with 98% accuracy, and produced a zenith angle distribution for real cosmic-muon events in excellent agreement with expectations.