Vertex AI: Pipelines for your MLOps workflows
summary
Shows a working Vertex AI pipeline that builds an image classifier from Cloud Storage, trains with AutoML or custom containers, then deploys to a REST endpoint; demo uses ~600 images per flower class and yields wrong labels when no outlier class is provided. Pipeline steps use cached steps so repeated runs take seconds while a full retrain takes ~13 minutes. Triggers use Cloud Scheduler, Cloud Build or Eventarc events; components shown include Feature Store, Metadata, batch prediction and explainable AI.
AI-written summary · about
chapters (6)
- 0:00Start
- 2:58Levels of automation defines maturity of the ML process
- 11:15What's included in Vertex Al?
- 14:01Where does Vertex Al fit in?
- 26:07Pipeline Components
- 29:31Developer friendly components
related talks
45mVertex AI: Pipelines for your MLOps workflows
31mVertex AI: Pipelines for your MLOps workflows
26mBuilding Portable ML Pipelines for Development and Production
12mBuild end-to-end solutions with Vertex AI
51mHands on with Cloud AI
30mAI in the Cloud
17mAI/ML demo derby
29mAI-Powered Anomaly Detection in Cloud Data Pipelines