Skip to content

usatlas-ml-training/Anomaly-Detection

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

50 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Anomaly detection using autoencoder

This repository is a fork of the original ADC2021 data challenge on anomaly detection at 40 MHz. This tutorial was modified to best suit the US ATLAS Machine Learning Training Event 2022. Accompanying introducing anomaly detection can be found directly at the link

This tutorial will cover:

  1. Examples of model architectures using autoencoders for anomaly detection
  2. How to train the model
  3. Compute the model predictions on background data
  4. Compute the model predictions on signal (anomalies)
  5. Evaluate model's performance using 2 different techniques

Autoencoder based on fully connected layers. Take a look at the file:

Dense_AE.ipynb

to follow along. Even though the architecture is pretty simple, we are still capable of finding all the anomalies with high background rejection! If you have time, change the basic model architecture and see if you can get an even better result!

Autoencoder based on convolutional neural networks. Take a look at the file:

Convolutional_AE.ipynb

to follow along. In this example, images are created using different particles as the y-coordinate and their features as the x-axis. Even though the images we create are not the standard image one would expect, we still get a pretty model also capable of identifying all anomalies! Can you make that model even better?

About

Contains code for Anomaly Detection hands-on section

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published