Bảng tổng hợp toàn khoá GCP Professional Machine Learning Engineer. GCP service reference, evaluation metrics, domain weights, và chiến lược thi.
Luyện thi Google Cloud Professional Machine Learning Engineer
Phần 5: Responsible AI & Ôn tập
xdev.asia
1. Cấu Trúc Đề Thi GCP Professional ML Engineer
Item
Details
Total Questions
60 câu
Time Limit
120 phút (2 giờ)
Passing Score
~70% (Google không công bố chính xác)
Format
Multiple choice, multiple select
Validity
2 năm
Level
Professional (intermediate to advanced)
2. Domain Weights
Domain
Weight
1. Architecting low-code ML solutions
~10%
2. Collaborate within and across teams to manage data and models
~20%
3. Scale prototypes into ML models
~20%
4. Serve and scale models
~20%
5. Automate & orchestrate ML pipelines
~20%
6. Monitor ML solutions
~10%
3. GCP ML Services Cheat Sheet
Task
GCP Service
No-code image classification
Vertex AI AutoML Image
SQL-based ML in data warehouse
BigQuery ML
Custom TensorFlow/PyTorch training
Vertex AI Custom Training
Hyperparameter optimization
Vertex AI Hyperparameter Tuning (Bayesian)
Feature consistency training/serving
Vertex AI Feature Store
ML workflow orchestration (pipelines)
Vertex AI Pipelines (KFP)
Experiment tracking
Vertex AI Experiments
Model versioning
Vertex AI Model Registry
A/B testing model versions
Vertex AI Endpoints traffic splitting
Monitor feature skew/drift
Vertex AI Model Monitoring
Explain model predictions
Vertex AI Explainability (SHAP, IG)
Real-time event ingestion
Pub/Sub
Batch + streaming ETL (unified)
Dataflow (Apache Beam)
Spark/Hadoop workloads
Dataproc
ML pipeline orchestration (multi-service)
Cloud Composer (Airflow)
Natural language analysis (no training)
Cloud Natural Language API
Document extraction
Document AI
Speech to text
Cloud Speech-to-Text API
Prevent data exfiltration
VPC Service Controls
Customer-managed encryption
Cloud KMS (CMEK)
4. Common Exam Traps
Trap
Correct Answer
"No ML expertise, image classification"
AutoML Image (not custom training)
"Train on data already in BigQuery"
BigQuery ML (not Vertex AI)
"Features differ at training vs serving"
Vertex AI Feature Store (not re-training)
"Trigger retraining when data arrives"
GCS notification → Eventarc → Vertex AI Pipeline
"Explain why model rejected application"
Vertex AI Explainability (SHAP)
"Train on distributed hospital data"
Federated Learning
"Prevent BigQuery data exfiltration"
VPC Service Controls
"Compare model performance across runs"
Vertex AI Experiments
Exam tip: GCP Professional ML Engineer thường hỏi về architecture decisions, không phải API syntax. Key question patterns: "which service BEST fits the requirement", "what is the FIRST step", "which approach requires the LEAST operational overhead". Luôn ưu tiên managed services của GCP khi câu hỏi có "minimal management" hoặc "serverless".
5. Kế Hoạch Ôn Tập
Ngày
Focus
Day 1
Vertex AI full platform: Training, Pipelines, Endpoints, Monitoring
Day 2
Data engineering: Pub/Sub, Dataflow, Dataproc, Cloud Composer