Federated ML for Edge Applications
summary
Fast Forward Labs at Cloudera describes federated ML that trains or partially trains models on-device and shares only parameter updates so raw text or images never leave phones, claiming improved global accuracy without centralizing private data. Architecture uses edge pre‑processing, intermediate regional nodes, and a central aggregator that averages model updates and applies privacy controls like differential privacy. Implementation relies on streaming and orchestration stacks such as Spark Streaming, messaging like Kafka, and Airflow plus containerized MLops to scale to thousands or hundreds of thousands of devices.
AI-written summary · about
chapters (13)
- 0:00Introduction and welcome
- 2:11Speaker background and career
- 5:00Role at Cloudera Fast Forward Labs
- 6:45Research process and multitask learning
- 11:45Federated ML for edge applications
- 21:14Edge device challenges and Fitbit experience
- 26:15Hierarchical ML environments
- 31:15Federated learning techniques
- 34:06Differential privacy and online learning
- 36:17Streaming and workflow standards
- 39:48AI ops and infrastructure challenges
- 45:04ML engineering and team roles
- 47:36Closing remarks
related talks
41mFederated Learning: Machine Learning on Decentralized Data
12mFederated Learning në ML/AI - mësimi i decentralizuar
34mDecentralized Machine Learning
44mFederated ML on iOS: Protecting User Data in the Age of AI
26mMachine Learning without the Data #federatedlearning
36mEdgeLake-FL: An Automated Federated Learning Platform for the Edge
04mProject Lightning Talk: Empowering Federated Learning with Multi-Cluster Management for Privacy and Efficiency
42mFederated Learning: Unlocking the Values of Discrete Data for Enterprise