Advertising, algorithms and Privacy: recent opensource developments
summary
Discusses concrete privacy wins and failures in advertising and algorithms: demonstrates reproducible de-anonymization of the Netflix Prize dataset, membership inference attacks that detect whether an individual was in a training set, and face/audio perturbation demos that leak sensitive traits. Reviews open-source differential privacy libraries from Google and Microsoft, explains local vs global differential privacy, privacy budgets and the loss of utility as epsilon tightens, and surveys browser rollouts (Safari/Firefox/Edge ITP) and adtech changes showing some defenses work in practice while many remain fragile.
AI-written summary · about
chapters (14)
- 0:00Talk overview
- 1:51Ad giants contributions
- 3:39Privacy risks in recommendations
- 5:06Facial recognition attacks
- 7:57Privacy budget risks
- 10:31Open source privacy tools
- 12:38Differential privacy explained
- 15:50Titanic dataset privacy
- 18:44Group by attacks
- 20:51Differential privacy in ML
- 23:24Ad ecosystem roles
- 25:11Privacy costs and regulations
- 26:54Browser privacy protection
- 27:57Ethical concerns
related talks
38mPrivacy: the Last Stand for Fair Algorithms
29mKeynote Presentation: Minding the (Privacy Research/Practice) Gap
27mThe ethical algorithm
32mPlatform-supported Auditing Of Social Media Algorithms For Public Interest
01h 10mDIGITAL PLATFORMS, THE NEW PRIVACY CHAMPIONS? BETWEEN MYTHS AND REALITIES
28mPrivacy Engineering Networking Break AMA
01h 20mUnlocking societal benefits of AI with privacy protective technologies.
33mPractical Data Privacy