Runtime Modifications of Distributed Big Data AI Models and Their Parameters
summary
Describes a system for hot-swapping AI models and their input parameters in distributed data pipelines by using a library of small packaged functions plus a controller that coordinates Spark clusters and Docker containers with Kafka-style ingestion. Demonstrates rolling updates and rollbacks on sensor-driven use cases (earthquake, vandalism), performing 12 live updates in tests with about 1% runtime overhead so updates apply without stopping jobs and while preserving pipeline state.
AI-written summary · about
chapters (6)
- 0:00Runtime modifications in distributed AI
- 6:22Real-time parameter changes
- 13:10Live parameter challenges
- 20:19Parameter update constraints
- 27:45Real-time data pipeline
- 33:07Q&A
related talks
16mOperationalizing AI in Finance: Real-Time Analytics
26mOptimal Approaches for Real-Time Machine Learning with Apache Spark on Kubernetes.
33mDealing with the versioning of production-ready models
27mDeveloping and Deploying Models at Scale — without Spark!
24mTaming Data Skew in Production ML Pipelines
53mData Engineering and AI
33mDeploying and Scaling Spark ML and Tensorflow AI Models
28mAI-driven query plan optimization in big data processing tools