AWS Certified Machine Learning Engineer - Associate (MLA-C01) Fact Sheet¶

Exam Overview¶

Exam Code: MLA-C01 Exam Name: AWS Certified Machine Learning Engineer - Associate Duration: 170 minutes (2 hours 50 minutes) Questions: 65 questions Question Format: Multiple choice and multiple response Passing Score: 720/1000 (scaled scoring, approximately 72%) Cost: $150 USD (50% discount for previous AWS certification holders) Valid For: 3 years Prerequisites: None required, but 1+ years experience with SageMaker recommended Language: Available in English, with more languages coming Delivery: Pearson VUE (online proctored or testing center) Launch Date: April 2025

📖 Official Exam Page - Registration and details 📖 Exam Guide PDF - Detailed exam objectives 📖 Sample Questions - Official practice questions

Target Audience¶

This certification is designed for: - ML engineers deploying and maintaining ML models in production - Data scientists moving into ML engineering roles - Software engineers working with ML pipelines - DevOps engineers managing ML infrastructure - MLOps professionals implementing ML workflows

Recommended Experience: - 1+ years working with Amazon SageMaker - Experience with ML model development and deployment - Understanding of MLOps practices - Familiarity with AWS infrastructure services - Python programming skills

📖 ML Engineering Learning Path - Official learning path 📖 Amazon SageMaker - ML platform overview

Exam Domains¶

Domain 1: Data Preparation for Machine Learning (28%)¶

This is the largest domain, covering data ingestion, transformation, and preparation.

1.1 Data Ingestion and Storage¶

Data Sources: - S3 for object storage - RDS and Redshift for structured data - DynamoDB for NoSQL data - Kinesis for streaming data - AWS Glue Data Catalog for metadata

📖 Amazon S3 - Object storage 📖 S3 for ML - ML data storage 📖 Amazon RDS - Relational databases 📖 Amazon Redshift - Data warehouse 📖 Amazon DynamoDB - NoSQL database 📖 Amazon Kinesis - Streaming data

1.2 Data Transformation with AWS Glue¶

AWS Glue Components: - Glue Data Catalog - Glue ETL jobs - Glue DataBrew for visual data prep - Glue crawlers for schema discovery - Glue Studio for workflow creation

📖 AWS Glue - ETL service 📖 Glue ETL - ETL jobs 📖 AWS Glue DataBrew - Visual data preparation 📖 Glue Crawlers - Schema discovery 📖 Glue Studio - Visual workflow

1.3 Data Preprocessing with SageMaker¶

SageMaker Data Wrangler: - Visual data exploration - Data transformation flows - Feature engineering - Export to SageMaker Processing or Pipelines

📖 SageMaker Data Wrangler - Data preparation 📖 Data Wrangler Transformations - Transform data 📖 Data Wrangler Export - Export options

SageMaker Processing: - Distributed data processing - Custom preprocessing scripts - Built-in containers (scikit-learn, pandas) - Integration with Spark

📖 SageMaker Processing - Data processing jobs 📖 Processing Containers - Using containers 📖 Spark Processing - Spark integration

1.4 Feature Engineering¶

Feature Store: - Online and offline feature stores - Feature groups and definitions - Feature versioning - Real-time and batch access

📖 SageMaker Feature Store - Feature management 📖 Feature Groups - Creating features 📖 Feature Store Access - Using features

Feature Engineering Techniques: - Encoding categorical variables - Scaling and normalization - Feature selection - Dimensionality reduction - Handling missing data

📖 Built-in Algorithms - SageMaker algorithms

1.5 Data Validation and Quality¶

Data Quality Checks: - Schema validation - Data profiling - Anomaly detection - Distribution checks - Missing value analysis

📖 SageMaker Model Monitor - Data quality monitoring 📖 Data Quality Monitoring - Quality metrics 📖 SageMaker Clarify - Data analysis

Domain 2: ML Model Development (26%)¶

Covers model selection, training, and tuning.

2.1 Selecting ML Approaches¶

Problem Types: - Supervised learning (classification, regression) - Unsupervised learning (clustering, anomaly detection) - Reinforcement learning - Deep learning approaches

📖 ML Concepts - ML fundamentals 📖 Algorithm Selection - Choosing algorithms

Built-in Algorithms: - XGBoost for tabular data - Linear Learner for classification/regression - Image Classification for computer vision - Object Detection (SSD) - Seq2Seq for NLP

📖 Built-in Algorithms - Algorithm reference 📖 XGBoost - Gradient boosting 📖 Linear Learner - Linear models 📖 Image Classification - Computer vision 📖 BlazingText - Text classification

2.2 Training Models with SageMaker¶

Training Job Configuration: - Choosing instance types - Distributed training strategies - Spot instances for cost optimization - Checkpointing and resuming - Input/output data configuration

📖 SageMaker Training - Training overview 📖 Training Jobs - Creating training jobs 📖 Distributed Training - Multi-node training 📖 Managed Spot Training - Cost optimization 📖 Checkpointing - Save training state

Custom Training: - Bring Your Own Container (BYOC) - Script mode with framework containers - Using TensorFlow, PyTorch, MXNet - Custom training scripts

📖 Bring Your Own Container - Custom containers 📖 TensorFlow on SageMaker - TensorFlow framework 📖 PyTorch on SageMaker - PyTorch framework 📖 Script Mode - Framework containers

2.3 Hyperparameter Tuning¶

SageMaker Automatic Model Tuning: - Hyperparameter search strategies - Bayesian optimization - Random search - Warm start tuning - Early stopping

📖 Hyperparameter Tuning - HPO overview 📖 Tuning Jobs - How it works 📖 Warm Start - Transfer learning 📖 Early Stopping - Stop poorly performing jobs

2.4 Model Evaluation¶

Evaluation Metrics: - Classification: Accuracy, precision, recall, F1, AUC-ROC - Regression: RMSE, MAE, R² - Clustering: Silhouette score, Davies-Bouldin - Custom metrics

📖 Model Evaluation - Model quality 📖 SageMaker Experiments - Experiment tracking 📖 SageMaker Debugger - Training insights

2.5 Model Versioning and Registry¶

SageMaker Model Registry: - Registering model versions - Model approval workflows - Model lineage tracking - Metadata management

📖 Model Registry - Model versioning 📖 Model Packages - Creating packages 📖 Model Approval - Approval workflows 📖 SageMaker Lineage - Track artifacts

Domain 3: Deployment and Orchestration (24%)¶

Covers model deployment, endpoints, and ML pipelines.

3.1 Model Deployment Options¶

Real-Time Inference: - SageMaker Endpoints - Multi-model endpoints - Multi-container endpoints - Serverless inference - Auto-scaling configuration

📖 SageMaker Inference - Deployment overview 📖 Real-Time Endpoints - Hosting models 📖 Multi-Model Endpoints - Multiple models 📖 Serverless Inference - Serverless deployment 📖 Auto Scaling - Scale endpoints

Batch Inference: - Batch Transform jobs - Large-scale predictions - Output configuration - Data filtering

📖 Batch Transform - Batch predictions 📖 Transform Jobs - How it works

Asynchronous Inference: - Long-running predictions - Queue-based invocations - Large payload support

📖 Async Inference - Asynchronous endpoints

3.2 ML Model Optimization¶

Model Optimization Techniques: - SageMaker Neo for compilation - Model quantization - Edge deployment with SageMaker Edge - Inference optimization

📖 SageMaker Neo - Model optimization 📖 Neo Compilation - Compile models 📖 Edge Manager - Edge deployment

3.3 ML Pipelines and Orchestration¶

SageMaker Pipelines: - Pipeline creation and execution - Pipeline steps (processing, training, tuning, etc.) - Conditional execution - Pipeline parameters - CI/CD integration

📖 SageMaker Pipelines - ML workflows 📖 Pipeline Steps - Defining steps 📖 Conditional Steps - Logic in pipelines 📖 Pipeline Parameters - Parameterization

AWS Step Functions for ML: - State machine orchestration - Parallel training jobs - Error handling and retries - SageMaker integration

📖 Step Functions - Workflow orchestration 📖 Step Functions ML - SageMaker integration

3.4 Infrastructure Management¶

Compute Selection: - Training instance types - Inference instance types - GPU instances (P3, P4, G4) - Inferentia instances (Inf1, Inf2) - Graviton instances

📖 Instance Types - Pricing and types 📖 GPU Instances - Accelerated computing 📖 AWS Inferentia - ML inference chips

Container Management: - SageMaker managed containers - Custom container registries (ECR) - Container optimization - Security scanning

📖 Amazon ECR - Container registry 📖 Container Security - Secure containers

3.5 CI/CD for ML¶

MLOps Automation: - CodePipeline for ML - CodeBuild for model building - CodeDeploy for model deployment - Integration with SageMaker Projects

📖 SageMaker Projects - MLOps templates 📖 MLOps Workflows - Project templates 📖 AWS CodePipeline - CI/CD automation

Domain 4: ML Solution Monitoring, Maintenance, and Security (22%)¶

Covers monitoring, security, and ongoing maintenance.

4.1 Model Monitoring¶

SageMaker Model Monitor: - Data quality monitoring - Model quality monitoring - Bias drift detection - Feature attribution drift - Real-time monitoring

📖 Model Monitor - Monitoring overview 📖 Data Quality - Data drift 📖 Model Quality - Performance drift 📖 Bias Drift - Fairness monitoring 📖 Capture Data - Endpoint data

4.2 Logging and Observability¶

CloudWatch Integration: - Training and endpoint metrics - Log aggregation - Alarms and notifications - Custom metrics

📖 CloudWatch Metrics - SageMaker metrics 📖 CloudWatch Logs - Log collection 📖 CloudWatch Alarms - Alerting

AWS X-Ray for ML: - Distributed tracing - Performance analysis - Bottleneck identification

📖 AWS X-Ray - Application tracing

4.3 Model Retraining and Updates¶

Retraining Strategies: - Scheduled retraining - Trigger-based retraining - Online learning approaches - A/B testing and canary deployments

📖 Model Retraining - Retraining pipelines 📖 Endpoint Updates - Safe deployment

Deployment Strategies: - Blue/green deployments - Canary deployments - Shadow deployments - Traffic shifting

📖 Deployment Guardrails - Safe updates 📖 Traffic Routing - Production variants

4.4 Security Best Practices¶

Data Security: - Encryption at rest (S3, EBS) - Encryption in transit (TLS) - AWS KMS for key management - VPC endpoints for private connectivity

📖 Security in SageMaker - Security overview 📖 Data Encryption - Encryption 📖 VPC Configuration - Private networking 📖 AWS KMS - Key management

IAM and Access Control: - IAM roles for SageMaker - Resource-based policies - Service Control Policies (SCPs) - Fine-grained permissions

📖 IAM for SageMaker - Access control 📖 IAM Roles - Execution roles 📖 Least Privilege - Policy examples

4.5 Compliance and Governance¶

Model Governance: - Model cards for documentation - SageMaker Model Registry - Audit trails with CloudTrail - Compliance certifications

📖 Model Cards - Model documentation 📖 CloudTrail Logging - API auditing 📖 Compliance - AWS compliance programs

Responsible AI: - SageMaker Clarify for bias detection - Explainability and interpretability - Fairness metrics - Human-in-the-loop workflows

📖 SageMaker Clarify - Bias and explainability 📖 Explainability - SHAP values 📖 Ground Truth - Data labeling

Study Strategy¶

Recommended Timeline (8-10 weeks, 12-18 hours/week)¶

Weeks 1-3: Data Preparation and Feature Engineering - Study data ingestion and storage - Learn AWS Glue and SageMaker Data Wrangler - Practice feature engineering - Work with Feature Store - Study time: 15 hours/week

Weeks 4-6: Model Development and Training - Study SageMaker built-in algorithms - Practice training jobs - Learn hyperparameter tuning - Work with model registry - Study time: 18 hours/week

Weeks 7-8: Deployment and Pipelines - Study endpoint deployment options - Learn SageMaker Pipelines - Practice CI/CD for ML - Build end-to-end workflows - Study time: 15 hours/week

Weeks 9-10: Monitoring, Security, and Review - Study Model Monitor - Learn security best practices - Take practice exams (aim for 75%+) - Review weak areas - Study time: 12-15 hours/week

Study Resources¶

Official AWS Training: 📖 AWS Skill Builder - Free AWS training 📖 ML Engineer Learning Plan - Official study plan 📖 Exam Prep Course - Official exam prep

Hands-On Practice: - Build end-to-end ML pipelines in SageMaker - Deploy models with different inference options - Set up Model Monitor for drift detection - Create MLOps workflows with SageMaker Projects - Practice with SageMaker Studio

📖 SageMaker Examples - GitHub examples 📖 SageMaker Immersion Day - Hands-on workshop

Exam Day Tips¶

Preparation¶

• Review SageMaker service limits and quotas
• Know endpoint deployment options
• Understand pipeline orchestration
• Review Model Monitor capabilities
• Get adequate rest before exam

During Exam¶

• Read questions carefully - focus on ML engineering aspects
• Look for keywords: "MOST cost-effective", "LEAST operational effort"
• Consider scalability and production requirements
• Eliminate wrong answers first
• Flag uncertain questions for review
• Manage time: ~2.6 minutes per question

Common Question Patterns¶

• Data preparation and feature engineering workflows
• Choosing appropriate training configurations
• Deployment strategy selection
• Pipeline orchestration and automation
• Monitoring and retraining strategies
• Security and access control
• Cost optimization approaches

Technical Setup (Online Proctoring)¶

• Stable internet connection
• Webcam and microphone required
• Clear workspace
• Government-issued photo ID
• Close all applications
• 170 minutes - plan breaks strategically

📖 Certification Preparation - Official resources

After Certification¶

Career Benefits¶

• Demonstrates ML engineering expertise
• Validates SageMaker proficiency
• Opens MLOps and ML Engineer roles
• Industry recognition for ML production skills

Next Certifications¶

📖 AWS Machine Learning Specialty - Advanced ML certification 📖 AWS Solutions Architect Professional - Architecture mastery 📖 AWS DevOps Engineer Professional - DevOps focus

Continuous Learning¶

• Follow AWS Machine Learning blog
• Experiment with new SageMaker features
• Attend re:Invent ML sessions
• Build production ML systems
• Join MLOps communities

📖 AWS ML Blog - Latest updates 📖 SageMaker Developer Guide - Complete documentation

Quick Reference¶

Exam Details at a Glance¶

• 65 questions in 170 minutes = ~2.6 minutes per question
• 720/1000 to pass = Approximately 72%
• 28% Data preparation = ~18 questions
• 26% Model development = ~17 questions
• 24% Deployment = ~16 questions
• 22% Monitoring & security = ~14 questions

Key SageMaker Components¶

Instance Type Selection¶

Good luck with your AWS Certified Machine Learning Engineer - Associate exam! 🎉