Machine Learning for CybersecuritySource on GitHub

Contents

Unit 12. Capstone project

The capstone is a project of your own: a security problem, data you are authorized to use, a model, and an evaluation that someone else could check and repeat. The single lesson of this unit gives the plan. It shows how to state the threat model and the decision the model supports, how to evaluate with a protocol that gathers the practices of the whole course (splits in time or by group, leakage checks, the base rate and the alert budget, simple baselines, intervals across seeds, per-family and per-group results, robustness, drift and ablations), how to write the report and the reproducibility package, and how to disclose real vulnerabilities responsibly. A rubric lets you assess your own work, and project ideas are offered at three sizes.

After the unit you will be able to plan a project whose conclusions follow from its evidence, recognize the common flaws of security machine learning studies in your own work and in published work, and deliver a report and code package that others can rerun. The unit draws on every earlier unit, in particular Units 1, 2, 7, 10 and 11. It has no lab; the exercises ask you to critique flawed project designs. Plan on about 2 hours for the lesson, exercises and quiz, plus 10, 25 or 50 hours for a small, medium or large project.

Lessons

  1. 12.1 Planning, running and reporting a project

Practice