Material for the ML tutorial held during ATLAS Italia 2022.
You can follow the tutorial in several way. Some way are interactive (you can run the code), some are static. On github and on colab you won't find the precomputed output.
- on cernbox (static)
- using Swan (interactive) (or https://swan-k8s.cern.ch/ if you have GPU access) -> Share -> Projects shared with me -> TutorialML-AtlasItalia2022).) You will find the project only if you are registered to this tutorial (and I haven't forget you)
- using colab: see links below (interactive)
- on github: see the lins below (static)
- on your laptop (instruction below)
Just some fun material about the present (2022) status of ML around the world
A super quick introduction to neural networks, tensorflow and keras
Images classification (fashion-mnist dataset) with a plain neural network
An improved version of the previous example using a convolutional neural network
Build an energy calibration for electron in ATLAS using a neural network: introduction to the problem and the dataset
The real neural network for the energy calibration
Some more complicated: build a regression of the distribution of the energy response of the detector to electrons
Use an autoencoder to generate fashion images
Use an autoencoder to remove noise from fashion images
Improve the generation of fashion images with a variational autoencoder
And make it conditional: generate your favourite fashion item
Reweight the MC to match the data using a NN correcting for several variables together.
Download the repository:
git clone [email protected]:wiso/TutorialML-AtlasItalia2022.git
you need a recent version of python (tested with python 3.10.4) and the possibility to install packages (here using a virtualenv, you can use conda instead)
cd TutorialML-AtlasItalia2022/
python -m virtualenv myenv --system-site-packages # the last option if you have ROOT already installed
source myenv/bin/activate
pip install -r requirements.txt
Open the first notebook:
cd notebooks
jupyter notebook 0.1-IntroML.ipynb