ML4Anomalies

Data samples used:

Can be found here.

Some useful references on the topic:

• On VAEs:

  • https://keras.io/examples/generative/vae/
  • https://www.tensorflow.org/guide/keras/custom_layers_and_models
  • https://towardsdatascience.com/understanding-variational-autoencoders-vaes-f70510919f73
  • https://arxiv.org/abs/1811.10276
  • https://arxiv.org/pdf/1906.02691.pdf

• On VAEs for anomaly detection in HEP:

  • https://arxiv.org/pdf/1312.6114.pdf

Some additional tests for anomaly detection:

Keras: Generic Keras setup and playground
In particular the folder VAE could be useful here

• https://github.com/amassiro/MyKeras

Anomaly detection for de/dx: Some tests for anomaly detection for de/dx based analysis

• https://github.com/amassiro/ML4Anomalies

Contents:

VAEmodel

Contains the VAE models used.

Training

Contains the script that runs the training of the VAE models.

Plots

Contains tools to plot input and output variables, latent variables, correlation matrix and ROC curves.

SHAP

Contains tools to understand which features are contributing the most to the loss of the model.

VAE and DNN:

Contains a new version of the model, comprised of a simple VAE together with a DNN that serves as a classifier for SM - EFT discrimination

Trained models:

vae_batch16_newModelDimenstions_MinMaxScaler_20_10_7_3_100
vae_test_newModelDimenstions_MinMaxScaler_20_10_7_3_100
vae_test_newModelDimenstions_MinMaxScaler_30_20_10_5_100