LorenzoValente3 / JointVAE4AD

Disentangle joint continous and discrete representations for Anomaly Detection in High Energy Physics.

JointVAE for Anomaly Detection

Anomaly Detection of HEP data done with Joint Variational Autoencoder.

The aim is to discriminate the top jets signals from the QCD background, but in an unsupervised setting (i.e. we only train on the QCD samples). The data being analyzed has been sourced from the git repo.

Project Structure

• ad: the main namespace.
  • models: defines custom models.
  • layers: defines custom layers used by the models.
  • metrics: custom metrics and wrappers.
  • evaluation: code for evaluating the models.
  • utils: general utility code.
  • plot: plot utils.
  • aug: data augmentations.
• weights: contains the pre-trained weights of the various models.