Becoming an AI Researcher
Milestone 2 of 4 · Research season · A year in the lab

Experiments that convince

Run the study so a sceptical reviewer would believe it: plan the comparisons, handle an exciting first result carefully, compute confidence intervals and paired tests, correct for multiple comparisons, build the ablation table and find test-set leakage before reviewers do. Includes 5 hands-on Python coding tasks (NumPy, SciPy, pandas), checked by automated tests.

Rigour50Communication50Ethics50Impact50Teamwork50
Month 2 · Mon 10:00The experiment planMonth 3 · Fri 17:30The results look greatMon 14:00Code it: seeds, means and confidence intervalsTue 10:00Code it: paired bootstrap, permutation and McNemarTue 15:00Code it: twelve datasets, correctedWed 11:00Code it: the ablation table, ready for LaTeXThu 10:00Code it: find leakage before reviewers doMonth 3 · Fri 10:00Telling the lab
Month 2 · Mon 10:00The experiment planYour desk
Experiment planBefore running anything
Kemi Adeyemi: Before you run a single job, write down what you will compare and how. Tick what belongs in the plan.

Every person, lab and university here is fictional. The decisions, documents and practices are modelled on how real AI research groups and PhD admissions work. The decisions, tools and practices are modelled on how real ML teams work.

Ready