Master IASD · DataLab 2026

Report guidelines

Back to the course

For every project

Your report is assessed on clarity, experimental rigor and the reasoning behind your conclusions. Expectations increase throughout the semester.

  • Five pages of content maximum, plus a group cover page. Exceeding the content limit incurs a penalty. Optional appendices are not read: keep essential evidence in the main report.
  • Keep the context brief. Focus on your choices, experiments and findings rather than retelling the history of the topic.
  • One method studied thoroughly can earn an excellent mark. A relevant baseline, targeted ablations and useful parameter studies matter more than a long list of methods.
  • Explain how methods were tuned and compared. One arbitrary configuration per method cannot establish which method is generally better; state what your experiments actually support.
  • Code itself is not graded. AI is allowed, but you must verify and be able to defend every claim. Strong leaderboard scores alone do not make a strong report; informative negative results are welcome.

A1 · Clarity and care

Present a clean, understandable experimental study.

  • Use readable equations, defined notation and figures with clear axes, legends, units where relevant and sensible numerical precision.
  • Explain the method actually used, the data splits, evaluation metrics and main settings.
  • Present results clearly and offer a first interpretation of what worked, what did not and why. A conventional method is perfectly acceptable.

A2 · Experimental analysis

Explain how your intervention affects the quality–diversity trade-off, while keeping the required generator architecture fixed.

  • Use controlled comparisons and targeted ablations to understand the effects of your choices.
  • If you run a hyperparameter search, analyse the trends. Explain sensitivity, plateaus, instability or trade-offs and what they suggest about the method. Listing the best configuration or describing which score increases is insufficient.
  • Interpret FID, precision, recall and generated images together. Connect observations to mechanisms, distinguish hypotheses from evidence and discuss failures.

A3 · A short research paper

Build a focused scientific argument, from a precise question to conclusions supported by evidence.

  • Define the evaluation setting precisely and justify fair, credible comparisons.
  • Use ablations and parameter analysis to test your explanations, consider alternatives and identify where the method succeeds or fails.
  • Discuss uncertainty, limitations and the scope of your conclusions. Aim for a concise, self-contained paper; a publishable discovery is not required.