2023/12/12 by Tejes Gaertner, Gaertner, Tejes, Jared Reiten +1 · 1 citation
Computer Science · Environmental Science · Physics and Astronomy · #Anomaly Detection Techniques and Applications #FOS: Physical sciences #High Energy Physics - Phenomenology (hep-ph) #Landslides and related hazards #Particle physics theoretical and experimental studies
paper · pdf · doi:10.48550/arxiv.2312.06948
openalex publication_date 2023/12/12 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
In the first part of this work, we demonstrate how the metric space structure induced by the energy mover's distance can be leveraged for the unsupervised tagging of jets according to their progenitor. Namely, we focus on the task of tagging jets initiated by a top quark from a background of jets initiated by light quarks and gluons. By examining the local neighborhood structure of this metric space, we find that the jets of each class populate the landscape in differing densities. This characteristic can be exploited to accurately cluster jets according to their densities through unsupervised clustering algorithms, such as DBSCAN. In the second part of this work, we modify the metric space by reducing the global notion of connectivity down to a local one and, in the process of doing so, modify our distance metric to be that corresponding to geodesics on an underlying graph. We demonstrate how this modification induces regions of both positive and negative values of curvature, which are then exacerbated through a Ricci flow algorithm. Differences in the curvatures averaged over local patches of the new graph metric space then lead to a flow which separates the signal top jets from the background in a fashion that is completely agnostic to any pre-determined jet labels.