Andreas Søgaard

**Suggested steps:** * [ ] Prepare tutorials / examples leveraging W&B for logging model training and results. * [ ] Better use the full functionality of W&B (logging training artifacts,...

**Suggested steps:** * [ ] Argue the importance of programming to physics departments * [ ] Facilitate collaboration between physics and computer science departments * [ ] Design and run...

Outline learning ressources for expected skills, tools, etc.

2

**Suggested steps:** * [ ] Create a dedicated document (e.g., markdown or wiki) related to learning resources. * [ ] Identify the technologies/tools used in GraphNeT. * [ ] For...

For instance, validation loss, accuracy, ROC curves, kinematic efficiency distributions etc. **Suggested steps:** * [ ] Implement classes / functions providing a default set of plots * [ ] Have...

This can be used to mitigate parametrised systematic uncertainties or data/Monte Carlo differences. Having the ability to run this form of training seamlessly would be a big benefit.

**Suggested formats:** * [ ] HDF5 * [ ] ROOT **NB:** Could be split up into individual issues for each format.

Use unsupervised learning to learn representations in data and Monte Carlo

7

**Suggested steps:** * [ ] Define unsupervised learning tasks, i.e., learning tasks that don't required truth-level labels but instead relies solely on the reconstruction-level data. This is the same principle...

Many of the tasks that the repo can do (data conversion, training, plotting, etc.) could well be exposed through a command-line interface (CLI). For instance this could take the form...

**Suggested steps:** * [ ] Use existing written tutorial(s) as baseline * [ ] Prepare content plan for various tutorial videos * [ ] Find place to host videos *...

Relates to #543