Learner Profiles
Who this lesson is for
- Workshop participants who have seen jets, quarks, or the Higgs boson discussed conceptually and want to see how those ideas turn into actual training data and code.
- Learners who are comfortable with basic Python but have not necessarily built or trained a neural network before.
Prerequisites
- Python: comfortable reading and writing basic Python (variables, functions, loops, and calling functions from a library). No prior experience with PyTorch, NumPy, or scientific Python libraries is assumed; this lesson introduces each one as it becomes needed.
- Machine learning: some general familiarity with the idea of training a model on labeled examples and then testing it on new data. No prior experience with neural networks or transformers is assumed. This lesson builds up every ML concept it uses (embeddings, self-attention, loss functions, training loops) from scratch, in plain language.
- Physics: no prior particle physics background is required. Jets, quarks, the Higgs boson, and how CMS records collisions are all introduced in the first two episodes.
- Account access: a Google account, used to run this lesson in Google Colab. No local software installation is required.
What this lesson does not assume
This lesson does not assume you have used Jupyter notebooks, Google Colab, or any CMS-specific software before. It also does not assume any familiarity with transformer architectures, since MiniParT is built and explained piece by piece starting from a plain neural network layer.