Domain 4: Machine Learning Implementation and Operations (20% of the exam content)

This domain accounts for 20% of the exam content.

Task 4.1: Build ML solutions for performance, availability, scalability, resiliency, and fault tolerance

• Log and monitor environments.

  • CloudTrail and Amazon CloudWatch

  • Build error monitoring solutions.

• Deploy to multiple Regions and multiple Availability Zones.

• Create AMIs and golden images.

• Create Docker containers.

• Deploy Auto Scaling groups.

• Rightsize resources (for example, instances, Provisioned IOPS, volumes).

• Perform load balancing.

• Follow best practices.

Task 4.2: Recommend and implement the appropriate ML services and features for a given problem

• ML on (application services), for example:

  • Amazon Polly

  • Amazon Lex

  • Amazon Transcribe

  • Amazon Q

• Understand service quotas.

• Determine when to build custom models and when to use Amazon SageMaker built-in algorithms.

• Understand infrastructure (for example, instance types) and cost considerations.

  • Use Spot Instances to train deep learning models by using Batch.

Task 4.3: Apply basic security practices to ML solutions

• Identity and Access Management (IAM)

• S3 bucket policies

• Security groups

• VPCs

• Encryption and anonymization

Task 4.4: Deploy and operationalize ML solutions

• Expose endpoints and interact with them.

• Understand ML models.

• Perform A/B testing.

• Retrain pipelines.

• Debug and troubleshoot ML models.

  • Detect and mitigate drops in performance.

  • Monitor performance of the model.