Privacy
Goals
- Analyze basic definitions of privacy
- Diagnose broken attempts to achieve privacy
- Reason about the implications of differential privacy
Conceptions of Privacy
- What are some basic conceptions of privacy?
- What are some limitations of each?
- What is contextual privacy?
- Compare/constrast informational vs. decisional privacy
- How can we achieve limited access and control?
- How have notions of privacy changed through history?
Early Attempts at Data Privacy
De-identification
- What are the two goals for privacy-preserving data disclosure?
- Why is removing user identifiers insufficient to preserve privacy?
- What are two forms of de-anonymization attacks?
- How did the Netflix-IMDB attack work?
- How does the support of an attribute affect its weight?
- Why do information amplification attacks work?
- How effective are the Netflix attacks?
- How effective is re-identification via linking?
K-Anonymity
- How is k-anonymity defined?
- How can a database be put into a k-anonymous form?
- What are three attacks on k-anonymity and how do they work?
- How does k-anonymity compose?
- Ex: If Alice is included in one database with X-anonymity and another
with Y-anonymity, what is her effective Z-anonymity level?
Differential Privacy
- What impossibility result did Dwork and Naor prove?
- What implications does it have for privacy-preserving data disclosure?
- Give an intuitive definition of differential privacy
- Give the formal definition of differential privacy
- What does the definition imply?
- How does differential privacy compose?
- Ex: If Alice is included in one database with X-DiffPriv and another
with Y-DiffPriv, what is her effective Z-DiffPriv level?
- How can DiffPriv be used in practice?
- How can DiffPriv be achieved in practice?
- Define function sensitivity
- Calculate function sensitivity for some exmaple database query functions
- What would it be for median? Max? Min?
- What is the Laplace mechanism?
- What does it imply about choices of function and epsilon?