Exploratory Data Science on Federated Raw Data
summary
A federated exploratory pipeline is built that ships algorithms to data sites instead of centralizing raw streams, using Apache SystemDS for parallelized model execution and NebulaStream for streaming; it runs on edge and cloud, handles a 70‑sensor anomaly detection (Gaussian mixture) and a ~100‑parameter paper‑quality regression, and scales with added workers—tested across a Denmark–Austria WAN test—showing performance competitive with scikit‑learn and TensorFlow.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:44Project EXTRA overview
- 1:28Siemens’ data science scope
- 3:11Federated learning approach
- 5:21Use cases in fertilizer and paper
- 9:16Federated system architecture
- 11:21Performance and scalability results
- 12:50Industries and applications
- 14:17Summary and conclusion
- 14:59Q&A
related talks
31mFrom Theory to Practice: Advances in Real-World Federated Learning
19mBig Data
34mLFN AI Taskforce: Platform-Driven Fede...
48mFederated ML for Edge Applications
48mPaths to anomaly detection w/ TIBCO data science, streaming on AWS
31mFederated Machine Learning with Kubernetes
25mData collaboration with Federated Learning
01h 26mData Driven Action: a primer on Data Science