📝 SummarXiv

Abstract

At the ATLAS and CMS experiments at CERN's Large Hadron Collider, the rate of proton-proton collisions far exceeds the rate at which data can be recorded. A real-time event selection process, or "trigger", is needed to ensure that the data recorded contains the highest possible discovery potential. In the absence of hoped-for anomalies that would lead to the discovery of new physics, there is increasing motivation to develop dedicated, model-agnostic, anomaly detection triggers. A common approach is to use unsupervised machine learning (ML) to predict an event-by-event anomaly score, based on the momenta and multiplicity of reconstructed objects. Such anomaly scores often exhibit high correlation with existing trigger observables and thus exhibit a selection bias towards high-momentum anomalies. In this article, we introduce DECorrelated Anomaly DEtection (DECADE), in which quantile regression is applied to the output of a pre-trained anomaly detection algorithm, guaranteeing the independence of the threshold on the anomaly score with respect to primary trigger observables. Thus, DECADE provides efficiency in low-momentum regions of phase space not captured by existing triggers, boosting the trigger efficiency for low-mass phenomena that are inaccessible via primary triggers and current anomaly detection triggers. Quantile regression is implemented using decision tree ensembles, making DECADE highly computationally efficient and therefore optimal for use both in software-based trigger systems and in FPGA-based hardware triggers. In both cases, we demonstrate that DECADE would add an insignificant additional latency and resource cost to the hardware anomaly detection triggers currently in operation at ATLAS and CMS, as well as to those proposed for the High-Luminosity era of the Large Hadron Collider.