Exercise on Anomaly Detection in Particle Physics for the 3rd Terascale School of Machine Learning.
Authors: Gregor Kasieczka, Louis Moureaux, Tobias Quadfasel and Manuel Sommerhalder (University of Hamburg).
- general introduction about anomaly detection, patricularly in a Particle Physics context
- Anomaly Detection using weak supervision methods
- Anomaly Detection with Autoencoders
There are two ways this exercise can be run:
- Using google Colab: In this case, no prior installation of Software is required. However, you need a google account to use the Colab service. For accessing the notebook, please follow this Link. Once you arrive at the notebook, click
File -> Save a copy in driveto get your personal copy of it, which you can then run and edit as you please. - Running locally: Of course, you can run this tutorial on your local computer or any other computing infrastructure you have access to. The notebook is located in this repository in the
exercise_anomaly_detection.ipynbfile. We provided anenvironment.ymlfile which you can use to install ananacondaenvironment that contains all necessary software packages. If you do not want to installanaconda, here is a list of recommendedpythonpackages that should be available on your machine in order to run the exercise:
pytorch(including the respective version ofcudatoolkitin case a GPU is available and should be used)pandaspytablesnumpymatplotlibscikit-learnh5pyvector
Note: It is highly recommended to run this exercise using a GPU if available. To use a GPU in Colab, klick Runtime->Change runtime type and choose GPU from the drop-down list under Hardware accelerator.
We also provide solutions to this exercise. These can be found in another Colab notebook under this Link, or as another ipynb notebook, which is located in the solution branch of this repository.