MLOps for edge devices with Amazon SageMaker Edge Manager
summary
SageMaker Edge Manager is discontinued April 26th, 2024 yet the session shows concrete MLOps: use SageMaker to train an autoencoder model, compile with SageMaker Neo, sign and package models, deploy to devices (Jetson Nano + Arduino sensors) via Greengrass or custom deploy, run a 2 MB agent serving multiple models, capture input/output samples to S3, monitor with SageMaker Model Monitor, relabel with SageMaker Ground Truth, retrain and redeploy — demo uses five 3D‑printed wind turbines and reports device-scale goals up to millions.
AI-written summary · about
chapters (8)
- 0:00Introduction to SageMaker Edge Manager
- 2:15Challenges of Edge ML operations
- 6:12Model optimization and serving on Edge
- 9:55Running multiple models efficiently
- 12:09Monitoring model quality decay
- 18:48Demo wind turbine use case
- 29:14Solution architecture overview
- 30:43Next steps and deployment guidance
related talks
37mStreamlining Machine Learning Operations
33mBuild, train, and deploy ML models using Amazon SageMaker
14mLinkEdge: Open-sourced MLOps Integration with IoT Edge
36mImplementing MLOps practices on AWS
34mImplement MLOps practices with Amazon SageMaker
34mBuild, train and deploy Machine Learning models with Amazon SageMaker
37mStreamlining Machine Learning Operations
19mUsing Amazon SageMaker to develop and deploy ML models