Sprecher
Oscar Marcos Perez Cytron
Beschreibung
We present an unsupervised machine learning approach to automate quality
monitoring in the HADES experiment, combining a Variational Autoencoder
with HDBSCAN clustering (VAE-HDBSCAN) to detect anomalies in detector
performance plots without needing to hand-label data. Our method
outperforms non-latent clustering and approaches the performance of
supervised CNN baselines, and is being integrated into Jefferson Lab's
HYDRA web platform for operator-friendly deployment during beamtime.