Schemes, Privacy, and Security
summary
Describes a practical DCGAN model inversion that reconstructs MRI training images from federated-model weights, measures cosine-similarity (many reconstructions 70–80% similar), and shows differential privacy reduces reconstruction quality at the cost of roughly 20% accuracy on complex image tasks. Surveys frameworks (TFF, FATE, IBM FL, OpenFL), evaluates aggregation algorithms, and demonstrates a fog node for IoT that lets resource-constrained devices join federated intrusion‑detection training with >90% system accuracy and only 1–5% drop when DP is applied.
AI-written summary · about
chapters (8)
- 0:00Federated learning concepts
- 5:25Open-source federated frameworks
- 12:07Personalized FL challenges
- 18:30Security and privacy in FL
- 26:34Privacy attacks and defenses
- 31:57Model inversion attacks
- 39:46FL with IoT devices
- 44:06Future directions
related talks
14mSoK: Data Reconstruction Attacks Against Machine Learning Models...
48mUnveiling the secrets of a machine learning model
31mFrom Theory to Practice: Advances in Real-World Federated Learning
26mOpenFL (Federated Learning) Building better AI models with private data
17mSoK: Gradient Inversion Attacks in Federated Learning
01h 12mModelling Privacy Threats of Medical Data and Health Data Spaces
32mInnovation in the Age of Regulation: Federated Learning with Flower
34mDecentralized Machine Learning