Privacy by Design Machine Learning
summary
Describes a privacy-by-design ML pipeline that moves training to user devices using federated learning, aggregates updates in Python notebooks with an open-source library, and protects model artifacts with AES-256 encryption plus HSM-backed key management. Demonstrates that naive models leak training examples via reconstruction attacks, measures the computation and network overhead of on-device training, and implements tested mitigations including secure aggregation and encrypted inference.
AI-written summary · about
chapters (8)
- 0:00Privacy by design for AI
- 3:00Data leaks and cloud risks
- 8:30Compliant AI technical solutions
- 15:00Federated learning decentralized models
- 22:00Hardware security encryption in ML
- 29:00Regulatory challenges compliance
- 36:00Future perspectives open-source
- 42:00Summary and Q&A
related talks
31mScaling privacy embedded distributed ML systems
19mPractical privacy-preserving machine learning in Python
29mBalancing Thin Line Between Data Intelligence & Privacy
43mGuarding the ML Galaxy: Beyond Accuracy to Privacy and Security
01h 28mPPML: Machine Learning on Data you cannot see
34mDecentralized Machine Learning
09mInterview with Blaise Aguéra y Arcas
28mPrivacy preserving machine learning