"Flux+Mutability": A Conditional Generative Approach to One-Class Classification and Anomaly Detection

Abstract

Anomaly Detection is becoming increasingly popular within the experimental physics community. At experiments such as the Large Hadron Collider, anomaly detection is at the forefront of finding new physics beyond the Standard Model. This paper details the implementation of a novel Machine Learning architecture, called Flux+Mutability, which combines cutting-edge conditional generative models with clustering algorithms. In the 'flux' stage we learn the distribution of a reference class. The 'mutability' stage at inference addresses if data significantly deviates from the reference class. We demonstrate the validity of our approach and its connection to multiple problems spanning from oneclass classification to anomaly detection. In particular, we apply our method to the isolation of neutral showers in an electromagnetic calorimeter and show its performance in detecting anomalous dijets events from standard QCD background. This approach limits assumptions on the reference sample and remains agnostic to the complementary class of objects of a given problem. We describe the possibility of dynamically generating a reference population and defining selection criteria via quantile cuts. Remarkably this flexible architecture can be deployed for a wide range of problems, and applications like multi-class classification or data quality control are left for further exploration.

Physics event classification using Large Language Models

“…These archived data were generated using the GEANT4 particle gun to create high-purity neutron and photon samples in the GlueX BCAL. The samples utilized the recon-2019_11-ver01.2 reconstruction version and 4.35.0 simulation version, and were constructed using the same conditions in Fanelli et al [5], e.g. a single shower condition to remove potential split-offs.1 The data were divided into two problems formulated as follows:…”

“…In each problem, separate training and test data sets were provided to participants. The data sets consisted of 14 feature variables describing the properties of electromagnetic showers [5]. The definitions of the characteristic variables are given in appendix A and include: the radial position of the shower, the energy deposited in each of the four BCAL layers, the energy fraction in each of the layers, and the widths of the shower in the z direction, radial direction, time dimension, azimuthal direction, and polar direction.…”

“…The shower profile of neutral particles, e.g., photons, and neutrons, is directly dependent on the phase space, characterized by the reconstructed 𝑧 position of a shower and the reconstructed energy 𝐸 [5]. In specific regions of the phase space, the interaction of neutrons within the calorimeter is limited, resulting in little energy deposition compared to photons and generally occurring in the outer layers.…”