Pretrained transformer finds new particle decays with 30x less data
bravo_abad · x · 2026-10-30
Pretraining meets particle physics search
Searching for new particles at colliders traditionally means training a separate model for each possible signal, with every search starting largely from scratch.
Hsu and authors built EVENET, a transformer pretrained on 500 million simulated collisions, combining:
- Diffusion-based reconstruction of hidden particle information
- Supervised tasks leveraging the simulator's knowledge of the underlying physics
Key result
Tested on a hypothetical Higgs decay absent from pretraining, the pretrained classifier — fine-tuned on just 10,000 examples — slightly exceeded the search sensitivity of the same architecture trained from scratch on 300,000 examples, a 30x reduction in required examples for the specific search.
The model still needs fine-tuning on new tasks, but learning known physics first transfers structural knowledge that dramatically cuts the data cost of new-physics searches.
More from Research
- Physicist calls OpenAI's Navier-Stokes solution physically meaningless despite loophole win — GaryMarcus · 2026-10-30
- ELLIS x UniReps talk: why distinct learning processes yield similar representations — ClementineDomi6 · 2026-10-30
- Anthropic paper: 0.1% pretraining poison plants RCE backdoor in 4B CLI agent — omarsar0 · 2026-10-30
- CopeNLU presents 7 papers at EMNLP 2026 on bias, alignment, explainability — IAugenstein · 2026-10-30
- Two-stage 3D ResNet pipeline detects neonatal brain pathologies from MRI with 0.919 AUC — Dr_Alex_Crimi · 2026-10-30
- 40 years of putting new neurons into the CNS: why so slow, and can AI accelerate it? — patrickmineault · 2026-10-30