Amazon SageMaker AI
- What it is
- The fully managed platform for the entire machine learning lifecycle: prepare data, build, train, tune, deploy, and monitor models.
- The middle tier: more control than the pre-built AI services, far less undifferentiated work than raw infrastructure.
- Key components
- SageMaker Studio — the web IDE for ML.
- Data Wrangler and Feature Store — data preparation and reusable feature management.
- Autopilot — AutoML: point it at tabular data and it builds, trains, and ranks models for you.
- JumpStart — pretrained models and solution templates ready to fine-tune. See Amazon SageMaker JumpStart.
- Ground Truth — managed data labeling.
- Managed training and endpoints — distributed training, automatic model tuning, real-time/serverless/batch inference.
- Model Monitor and Clarify — detect data drift, bias, and explain predictions.
- Where it sits in the three tiers
- Tier 1: prebuilt AI services (Rekognition, Comprehend) — no ML skills.
- Tier 2: SageMaker — you own the model, AWS owns the infrastructure.
- Tier 3: ML Frameworks on AWS ML Infrastructure — you own everything.