What, Why and How of Federated Machine Learning – Implementation Using...
summary
Deploys a two‑party FATE cluster on Kubernetes with KubeFATE and a FATE Operator for Kubeflow, trains a homo logistic regression on a Kaggle credit‑card fraud dataset split between parties, and exchanges only encrypted gradient updates. The demo uses a Jupyter notebook and the FML Site Portal to run the pipeline, publishes the federated model to FATE serving, converts to ONNX/TFLite, and reports model accuracy comparable to central training while calling out operational challenges around cluster management, isolation, and availability.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:18What, Why and How of Federated Machine Learning
- 2:32Federated Learning Introduction
- 3:45Federated Learning - Practical Solution to Data Availability (Silos)
- 5:21Federated Learning System Design (Mobile)
- 6:08Federated Learning Categories
- 7:47Federated Learning Concept
- 8:23Federated Learning Challenges
- 9:09Federated Al Technology Enabler (FATE)
- 10:34KubeFATE: FATE Cluster Management
- 12:37"Low-code" FL Platform
- 14:11Demo
- 19:50Resources
related talks
31mFederated Machine Learning with Kubernetes
42mFederated Learning: Unlocking the Values of Discrete Data for Enterprise- 31mFederated Machine Learning with Kubernetes
30mAccelerate Federated Learning Model Deployment with KServe (KFServing)
04mProject Lightning Talk: Empowering Federated Learning with Multi-Cluster Management for Privacy and Efficiency
26mDecentralized Federated Machine Learning: Empowering Edge Devices with Kubernetes
27mFederated Learning with Azure Machine Learning NVIDIA FLARE and MONAI
25mData collaboration with Federated Learning