Choose the reader closest to your experience for a suggested starting point.
No lab experience yet, and unsure whether to ask to join one now or show up with something in hand. Show up with something in hand — it takes about a month.
Read Unit 01 and write the study brief in its lab. That brief is the something.
Keep the dictionary open beside it. Much of what feels like difficulty here is vocabulary, and it is fixable in a week.
Work modules 01–04 for the framing the units assume you already have.
Read the hidden curriculum before you need it. It states the norms nobody says out loud — how to ask a question, what "read the paper" actually means.
You have the fundamentals and now need a defensible result. The trap is treating proofreading as somebody else's job: a model trained on annotations nobody checked learns the annotator's mistakes at scale.
Do Units 05–07 with the drills. This is the slowest part of the course and the one worth the most.
Then Unit 08, where the error taxonomy stops being abstract.
Work Research in Action for the claim-to-evidence discipline a defensible result needs.
You are also mentoring someone. The facilitator guide is the difference between mentoring by improvisation and mentoring by design.
You can build the model but not yet judge whether the data supports the claim. You do not need another machine learning course; you need the biology that makes the labels mean something.
Start at Unit 04 or Unit 08, where systems intuitions transfer directly.
Then go back to Units 05–07. Entry-point order is negotiable; skipping these is not.
