Heidelberg 2022 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 93: Search for New Particles 6

T 93.3: Vortrag

Donnerstag, 24. März 2022, 16:45–17:00, T-H24

Autoencoders and k-Means for unsupervised anomaly detection in high energy physics — Thorben Finke, Michael Krämer, Alessandro Morandini, Alexander Mück, and •Ivan Oleksiyuk — Institute for Theoretical Particle Physics and Cosmology (TTK), RWTH Aachen University, D-52056 Aachen, Germany

Unsupervised anomalous jet tagging based on low-level observables has recently gained popularity in the high energy physics community. The main goal here is to be as efficient and model-independent as possible. We scrutinize a widely used anomaly detection method based on the reconstruction loss of a deep autoencoder to show its capabilities, but also its limitations. Although we reproduce the positive results from the literature, we show that the standard autoencoder setup cannot be considered as a model-independent anomaly tagger by inverting the task: the autoencoder fails to tag QCD jets if it is trained on top jets. We improve the capability of the autoencoder to learn non-trivial features of jet images, such that it is able to achieve both top jet tagging and QCD jet tagging with the same setup. We propose an alternative machine learning approach using k-Means and Gaussian Mixture Model to construct anomaly scores. We show that these methods, albeit simple, have several benefits and may also be regarded as promising anomaly detection tools.