Machine Learning without the Data #federatedlearning
summary
Claims that practical privacy-preserving ML is possible using current tools and shows a working federated setup: uses the Flower framework with four hospitals and the binomia MNIST chest x‑ray dataset (~6,000 images), server runs Federated Averaging, server waits for 4 clients but samples 2 for training, clients train locally 50 epochs (local accuracy rises ~63%→90%, eval ~81%), aggregated loss ~0.46. Notes homomorphic encryption is currently impractical and differential privacy trades accuracy for noise.
AI-written summary · about
chapters (6)
- 0:00Start
- 1:51Transparency
- 3:14Centrally storing private data
- 5:07Homomorphic Encryption
- 7:36Differential Privacy
- 8:32Federated Learning
related talks
32mInnovation in the Age of Regulation: Federated Learning with Flower
18mEnabling Medical AI through Federated Learning
28mA Privacy Enhancing Federated Learning Framework for Multi-Party Distributed ..
42mLearning Together, Distributed: An Introduction to Federated Learning
40mDemystifying Decentralized Fair Models for Federated Machine Learning
34mDecentralized Machine Learning
41mFederated Learning: Machine Learning on Decentralized Data
48mFederated ML for Edge Applications