Challenges for unsupervised anomaly detection in particle physics

Abstract

Anomaly detection relies on designing a score to determine whether a particular event is uncharacteristic of a given background distribution. One way to define a score is to use autoencoders, which rely on the ability to reconstruct certain types of data (background) but not others (signals). In this paper, we study some challenges associated with variational autoencoders, such as the dependence on hyperparameters and the metric used, in the context of anomalous signal (top and W) jets in a QCD background. We find that the hyperparameter choices strongly affect the network performance and that the optimal parameters for one signal are non-optimal for another. In exploring the networks, we uncover a connection between the latent space of a variational autoencoder trained using mean-squared-error and the optimal transport distances within the dataset. We then show that optimal transport distances to representative events in the background dataset can be used directly for anomaly detection, with performance comparable to the autoencoders. Whether using autoencoders or optimal transport distances for anomaly detection, we find that the choices that best represent the background are not necessarily best for signal identification. These challenges with unsupervised anomaly detection bolster the case for additional exploration of semi-supervised or alternative approaches.

A semi-supervised approach to dark matter searches in direct detection data with machine learning

“…Anomaly detection refers to the practice of calculating a well defined 'anomaly score' in order to determine if an event or group of events are significantly discrepant from elements of the background distribution [54]. In statistical language, this process involves studying the asymptotic properties of the anomaly function T S(x, y), which is generally a function of the image data inputs x and VAE output y.…”

Quantum Anomaly Detection for Collider Physics

“…[16] in the context of autoencoders. These unsupervised tools can be trained without any simulation, but are not as effective as semi-supervised methods when there is a good background model and/or when the new physics is not the lowest density events [64,80]. For this reason, our focus is on semi-supervised learning.…”

Simulation-based Anomaly Detection for Multileptons at the LHC

“…Recently, there have been many proposals for automating AD methods with machine learning [62][63][64][65][66][67][68][69][70][71][72][73][74][75][76][77][78][79][80][81] (see Ref. [80][81][82][83] for overviews of the field).…”