Joint Variational Auto-Encoder for Anomaly Detection in High Energy Physics

Despite providing invaluable data in the field of High Energy Physics, the LHC may encounter challenges in obtaining interesting results through conventional methods applied in previous run periods. Our proposed approach involves a Joint Variational Autoencoder (JointVAE) model, trained on known physics processes to identify anomalous events corresponding to previously unidentified physics signatures. By doing so, this method does not rely on any specific new physics signatures, and it can detect anomaly events in an unsupervised manner, complementing the traditional LHC search tactics relying on model-dependent hypothesis testing. This paper also presents a study on the implementation feasibility of the JointVAE model for real-time anomaly detection in general-purpose experiments at CERN LHC. Low latency and reduced resource consumption are among the challenges faced when implementing machine learning models in fast applications, such as the trigger system of the LHC experiments. Therefore, the JointVAE model has been studied for its implementation feasibility in Field-Programmable Gate Arrays (FPGAs), utilizing a tool based on High-Level Synthesis named HLS4ML.