The Dangerous Data Anonymization - Keynote
summary
Data anonymization often fails in practice: simple masking with asterisks is reversible by joining multiple masked datasets, Netflix/IMDb-style linkage is cited, and a Washington medical release lets researchers reidentify ~43% of patients by cross‑referencing other public records. Techniques surveyed include k‑anonymity and l‑diversity clustering, UID substitution, aggregation (Strava heatmaps exposing secret bases), and differential privacy noise—each trades privacy for utility. Practical fix: run automated image redaction (AWS Rekognition was proposed ~\$1k/year) before publication.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:05What is anonymization
- 4:37Why anonymize data
- 8:57Basic anonymization techniques
- 16:48Risks of insufficient masking
- 19:43Aggregation exposure risks
- 22:35Pseudonymization dangers
- 25:52Mixed data responsibilities
- 30:29Summary and recommendations
related talks
10mHealth Data Anonymization and Privacy
29mPersonal Data Pseudonymization Versus Anonymization In The Age Of AI & Big Data
01h 17mTo be or not to be (anonymous)? Anonymyity in the age of big and open data.
15mHow We Can Save Anonymization
33mPractical Data Privacy
50mData Privacy Concerns - A Risk Based Approach to De-Identification
09mPrivacy Analytics and De-Identification
20mShare data without revealing personal information