TU Wien:Security, Privacy and Explainability in Machine Learning/2025-10-10 Exam-2025S

1. Name quality characteristics of good explanations and discuss
2. You are a hospital, datatset needs to stay local. what methods can you use to perform meaningful collaborative statisticl computation (e.g. linear regression)
3. Commonalities and differences between model watermarking and data exfiltration
4. Model stealing
   1. name threat models
   2. name attacks
   3. name two defences
5. face recognition: database of employees, what are potential security or privacy vulnerabilities