Dr. Alexander Mück

I will introduce some machine learning concepts for more or less model agnostic new physics searches at the LHC. Since autoencoders are a standard tool for anomaly detection, they have also been introduced in high energy physics as a promising architecture for model-independent BSM searches. I will introduce the concept of unsupervised anomaly detection based on the autoencoder reconstruction loss, show its capabilities, but also its limitations. Weakly supervised methods are an attractive alternative. I will introduce the concept of classification without labels (CWoLa) and show a physics application improving a mono-jet search.