Local Diagnostics of Learned Event Representations for Anomaly Detection at the HL-LHC
Supervised embeddings improve HL-LHC anomaly detection on 750 million simulated events and add local diagnostics.
The paper studies anomaly detection in learned event representations for High-Luminosity LHC new-physics searches on COLLIDE-2V, a 750-million-event simulation. Supervised neural embeddings generally outperform self-supervised ones and, in the best setups, improve on an autoencoder trained on kinematic features, raising signal efficiency at 10% background acceptance. A Gaussian Mixture Model diagnostic localizes flagged events and ranks high-level observables by local discriminating power, recovering the defining kinematics of held-out processes.
- Evaluated embeddings on COLLIDE-2V, a 750-million-event HL-LHC simulation.
- Supervised objectives generally beat self-supervised ones and a kinematic autoencoder.
- Best setups raise signal efficiency at 10% background acceptance.
- A GMM diagnostic ranks observables by local discriminating power.
From the original article
1 min read at arxiv.orgAnomaly detection, the identification of collision events that deviate from Standard Model expectations, is one of the most promising model-agnostic strategies for new-physics searches at the Large Hadron Collider.
By Donatella Genovese · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.