CDL
Deep Dive
Google Cloud's end-to-end platform for building, deploying, and managing ML models. From data to production.
Vertex AI is Google Cloud's unified ML platform (since April 2026 it is part of Gemini Enterprise Agent Platform; the CDL exam guide still uses the Vertex AI name). Before 2021, Google had dozens of separate ML services (AI Platform, AutoML Vision, AutoML NL, etc.) — Vertex AI unifies them into a single coherent platform.
It covers the entire ML lifecycle: data → experimentation → training → evaluation → deployment → monitoring.
One place
The entire ML lifecycle on one platform. Datasets, models, endpoints, pipelines, and Feature Store in one console.
AutoML + Custom
From no-code AutoML to training with TensorFlow/PyTorch. Choose based on your team.
Native MLOps
Reproducible pipelines, model versioning, drift monitoring, and CI/CD for ML.
GCP Hardware
Access to GPUs (NVIDIA A100, H100) and TPUs from Google for training at scale.
| Component | Function | When to use |
|---|---|---|
| Vertex AI Workbench | Managed Jupyter notebooks with GPU/TPU access. ML development environment in the cloud. | Data exploration and ML experimentation |
| AutoML | Trains classification, regression, vision, or NLP models with your data, no code required. | No ML team but you have labeled data |
| Custom Training | Runs your own code (TF, PyTorch, scikit) on managed hardware. Maximum control. | Data scientists with their own models |
| Model Registry | Centralized model repository with versioning, metadata, and lineage. | Managing multiple model versions in production |
| Vertex AI Endpoints | Deploys models as REST APIs for real-time inference. Auto-scaling. | Serving predictions to production applications |
| Batch Prediction | Runs predictions over a full dataset without a dedicated endpoint. | Offline predictions over millions of records |
| Feature Store | Centralized repository of ML features to share across teams and models. | Multiple models sharing the same data features |
| Vertex AI Pipelines | Orchestrates reproducible ML workflows using Kubeflow Pipelines or TFX. | MLOps: automated training and deployment pipelines |
| Experiments | Tracks and compares ML experiments: hyperparameters, metrics, artifacts. | Systematically comparing dozens of training runs |
| Model Monitoring | Detects data and prediction drift for models in production. | Ensuring production models don't degrade in performance |
AutoML tabular
Structured tabular data (CSV, BigQuery). Binary or multiclass classification, or regression. Automates feature engineering and model architecture.
Example use case
Predict which customers are going to cancel their subscription (churn).
AutoML image
Image classification, object detection, or segmentation. Upload labeled images and AutoML finds the best neural network architecture.
Example use case
Detect defects in manufactured parts using photos from the production line.
Text: Gemini + fine-tuning
AutoML text was retired in 2024: today, text classification, entity extraction, or sentiment with your own categories is handled with Gemini (prompts or fine-tuning).
Example use case
Classify support tickets by category (billing, technical, general) using ticket history.
AutoML video
Video classification, action recognition, or object tracking in video.
Example use case
Recognize exercises in fitness videos for a training app.
TensorFlow
Open-source ML framework created by Google in 2015. One of the most popular in the world for deep learning. Vertex AI Custom Training supports it natively.
Cloud TPU (Tensor Processing Unit)
A chip designed by Google specifically to accelerate ML operations (matrix multiplications). Available in Vertex AI and as VMs in Compute Engine.
GPU vs TPU: when to use each
GPU (NVIDIA A100, H100)
TPU (Cloud TPU)
Did you get this topic?
Test what you just learned
A data science team has their own TensorFlow models that need to be trained weekly with new data. They want to version the models, compare them, and automatically deploy the best one. Which Vertex AI components do they need?
Put what you learned to the test with original CDL exam questions.
Start CDL mock examUnlimited mock exams with the Pro plan
Log in to track your progress.