Privacy Challenges in Machine Learning
summary
Survey of concrete solutions: lattice-based cryptography (notably learning with errors) underlies NIST’s post‑quantum proposals and enables homomorphic and functional encryption; zero-knowledge proofs power privacy in cryptocurrencies and proposals for court and surveillance audits; practical deployments use multi-party computation for Danish sugar‑beet auctions, Estonian VAT checks and proposed satellite‑collision avoidance; recent papers report feasible secure neural‑net inference and federated‑learning pipelines, with real implementations showing performance tradeoffs but proving the core claims viable.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 5:27Basic research
- 9:55RSA
- 16:08Learning with Errors
- 18:30PostQuantum
- 19:06Zero Knowledge
- 23:17Standardization
- 23:59Blockchains
- 24:49Proof of Work
- 25:59Timestamping
- 27:35Applications
- 29:17Utility
- 30:12Math
- 30:47Multiparty Computation
- 31:32Application Examples
- 36:13Federated Learning
- 37:25Public Good
- 39:04Educational Outcomes
- 40:02Surveillance
- 41:05International Cooperation
related talks
29mPrivacy-preserving methods: Building secure projects
22mPrivacy-preserving Computation in Plain Sight
28mSecurity of Machine Learning
12mThe Limits of Technology in Achieving Privacy
26mDoes using data mean giving up privacy?
28mPrivacy preserving machine learning
43mGuarding the ML Galaxy: Beyond Accuracy to Privacy and Security
01h 20mUnlocking societal benefits of AI with privacy protective technologies.