MLA-C01（日语） 练习题 — MLA-C01:机器学习工程师-助理级（日语）

The scenario specifies a SageMaker-based ML workflow requiring a central model registry for version management, secure isolation of training data across the ML lifecycle, and minimal operational overhead. The suggested answer leverages native, fully managed SageMaker functionality purpose-built for this exact use case, eliminating the need for custom tooling or manual infrastructure management. SageMaker Model Registry natively integrates with all other SageMaker features including training jobs, IAM access controls for S3 training data, model deployment, and model monitoring, while model groups provide out-of-the-box version cataloging for related models, aligning perfectly with all requirements with the lowest possible operational effort. Option Analysis: A. Incorrect. Amazon ECR is a container image registry, not a purpose-built ML model registry, and does not natively store model metadata, training lineage, or support ML-specific approval workflows. Creating separate ECR repositories per model requires manual management of permissions, version tracking, and custom integration with SageMaker workflows, introducing significant unnecessary operational overhead and failing to deliver a native central model registry for the ML lifecycle. B. Incorrect. Using ECR with unique tags for model versions still relies on a container registry rather than a purpose-built ML model registry. This approach requires custom scripting to track model metadata, lineage, and integrate with SageMaker deployment and monitoring features, leading to far higher operational overhead than native SageMaker Model Registry functionality. ECR is not designed to serve as a central ML model registry, so this option does not meet the core requirement. C. Correct. SageMaker Model Registry is the fully managed, native central model registry for SageMaker ML workflows, designed explicitly to support end-to-end ML lifecycle management. Model groups within SageMaker Model Registry are purpose-built to catalog and organize all versions of a single model, automatically track version numbering, training data lineage, approval status, and integrate natively with SageMaker IAM controls to enforce secure, isolated access to S3 training data. This solution requires no custom infrastructure or manual management, delivering the lowest operational overhead while meeting all stated requirements. D. Incorrect. While SageMaker Model Registry is the correct service for this use case, using unique tags to manage model versions is inefficient and introduces unnecessary operational overhead. Model groups are a native built-in feature that automatically manages versioning, organization, and lineage for related model versions, eliminating the need for manual tag management, risk of tag collisions, and custom processes to track version order. This option is less efficient than option C. Key Concepts: 1. SageMaker Model Registry: A fully managed AWS service purpose-built for cataloging ML models, managing model versions, tracking lineage and metadata, and implementing deployment approval workflows, with native integration across the entire SageMaker ML lifecycle. 2. SageMaker Model Groups: A core construct within SageMaker Model Registry that groups all versions of a single use case model, automates version numbering, and simplifies organization of related model artifacts, metadata, and deployment history without custom configuration. 3. ML Operational Efficiency: A core MLA-C01 domain concept that prioritizes use of native integrated AWS ML services over custom disjointed tools to reduce operational overhead, improve security, and simplify end-to-end ML lifecycle management. References: Amazon SageMaker Model Registry, https://docs.aws.amazon.com/sagemaker/latest/dg/model-registry.html Organize Model Versions with Model Groups, https://docs.aws.amazon.com/sagemaker/latest/dg/model-registry-model-group.html

The question requires minimizing infrastructure startup times for consecutive SageMaker training jobs, while maintaining secure isolated use of training data. By default, each SageMaker training job provisions new compute instances, pulls the training container image, and initializes the runtime environment, a process that typically takes several minutes per job. For consecutive experimental training runs, this startup latency adds significant overhead. SageMaker managed warm pools address this directly by retaining the compute instances, container runtime, and associated environment configuration after a training job completes. Subsequent consecutive jobs can reuse these pre-provisioned, isolated resources instead of spinning up new infrastructure, reducing startup time from minutes to seconds. Warm pools are dedicated to the customer AWS account, so they maintain the required security and isolation for training data access to S3, aligning with all requirements in the scenario. This topic falls under the MLA-C01 domain of optimizing ML workloads on AWS, specifically performance optimization for SageMaker training. Option Analysis: A. Managed Spot Training is incorrect. This feature is designed to reduce training costs by leveraging unused EC2 capacity at discounted rates, not to reduce infrastructure startup time. Spot instances may have longer provisioning wait times if capacity is constrained, and can be interrupted, making them unsuitable for minimizing startup latency for consecutive jobs. B. SageMaker managed warm pools is correct. This feature retains pre-provisioned, isolated compute resources and runtime environments between consecutive training jobs, eliminating the overhead of resource provisioning, container pull, and environment initialization for subsequent runs. This directly minimizes infrastructure startup time as required, and maintains the required security isolation as warm pool resources are not shared with other AWS customers. C. SageMaker Training Compiler is incorrect. This tool optimizes training code execution to reduce the runtime of the training process itself, but has no impact on the infrastructure startup time that occurs before the training code begins running. It addresses a different phase of the training lifecycle than the requirement specified. D. SageMaker distributed data parallelism (SMDDP) library is incorrect. This library optimizes data transfer across multiple compute nodes for distributed training workloads, reducing training execution time for large, multi-node jobs. It does not affect the infrastructure startup time for individual consecutive training jobs, so it does not meet the requirement. Key Concepts: 1. SageMaker Warm Pools for Training: A core performance optimization feature for iterative and consecutive training workloads that retains provisioned compute resources between jobs to eliminate startup latency, a key topic in the MLA-C01 domain of ML workload optimization. 2. SageMaker Training Job Lifecycle: MLA-C01 candidates are tested on their knowledge of the distinct phases of a SageMaker training job (resource provisioning, initialization, execution, teardown) to identify targeted optimizations for specific workload pain points. 3. SageMaker Training Optimization Differentiation: Candidates must be able to distinguish between optimizations for cost (e.g., Managed Spot Training), training execution speed (e.g., Training Compiler, SMDDP), and infrastructure startup latency (e.g., Warm Pools) to select the right tool for a given use case. References: Use Warm Pools for Training, https://docs.aws.amazon.com/sagemaker/latest/dg/train-warm-pools.html AWS Certified Machine Learning – Associate (MLA-C01) Exam Guide

This question aligns with the MLOps and ML Governance domains of the AWS Certified Machine Learning Engineer - Associate (MLA-C01) exam, which test knowledge of orchestrating governed, repeatable ML workflows and enforcing pre-deployment control policies. The scenario requires two core capabilities: an end-to-end ML lifecycle solution covering experimentation, training, model registry, deployment and monitoring, plus a manual approval gate to block unvetted models from production deployment. SageMaker Pipelines is the purpose-built, fully managed CI/CD orchestration service for SageMaker ML workflows, and it natively integrates with the SageMaker Model Registry. The Model Registry includes built-in approval status fields for all model versions: PendingManualApproval, Approved, and Rejected. When a model is registered as part of a pipeline run, teams can configure the pipeline to pause for manual review, then use the AWS SDK to update the model version's status to Approved once review is complete. Deployment steps in the pipeline can be restricted to only run for Approved model versions, directly meeting the manual approval requirement. SageMaker Pipelines also supports the remaining required features in the scenario, including integration with SageMaker Experiments for experimentation tracking, isolated training job configuration for secure S3 data access, and integration with SageMaker Model Monitor for post-deployment monitoring. Option Analysis: A. Incorrect. SageMaker Experiments is a service for tracking, organizing, and comparing ML experiment runs, including metrics, hyperparameters, and run artifacts. It does not include native functionality to implement or facilitate manual approval workflows for model registration, so it cannot meet the core governance requirement in the scenario. B. Incorrect. SageMaker ML Lineage Tracking is used to track the origin and relationships of ML artifacts, including training data sources, training jobs, model versions, and deployed endpoints, for audit and compliance purposes. It does not provide approval workflow controls or status management for model registry entries, so it cannot support the required manual approval process. C. Incorrect. SageMaker Model Monitor is a post-deployment service used to detect data drift, concept drift, model quality degradation, and bias in production endpoints. It operates only after a model is deployed, so it cannot manage pre-deployment manual approval workflows for new model versions. D. Correct. SageMaker Pipelines orchestrates the entire ML lifecycle covered in the scenario, and natively integrates with the SageMaker Model Registry. Teams can configure new model versions to be registered with a PendingManualApproval status by default. After manual review is complete, the AWS SDK for SageMaker can be used to update the model version's status to Approved. Deployment steps in the pipeline can be configured to only execute for Approved model versions, ensuring no unvetted models are deployed to production. This solution meets all requirements in the scenario. Key Concepts: 1. SageMaker Model Registry Approval Controls: The SageMaker Model Registry includes built-in approval status tracking for all registered model versions, enabling teams to implement formal manual review processes before production deployment. Deployment workflows can be restricted to only use Approved model versions to enforce governance policies. 2. SageMaker Pipelines Orchestration: SageMaker Pipelines is a fully managed CI/CD service for ML workflows that integrates natively with all SageMaker ML lifecycle services, including Experiments, Training, Model Registry, and Model Monitor, to create governed, repeatable end-to-end workflows. 3. ML Workflow Governance: MLOps governance requires pre-deployment control gates such as manual approval to ensure only validated, compliant models are deployed to production, paired with access controls for sensitive training data to meet security requirements. References: Use the SageMaker Model Registry, https://docs.aws.amazon.com/sagemaker/latest/dg/model-registry.html Build and Manage SageMaker Pipelines, https://docs.aws.amazon.com/sagemaker/latest/dg/pipelines-build.html

The scenario requires an on-demand, automated workflow to monitor bias drift for SageMaker real-time deployed models, while maintaining secure, isolated access to training data stored in S3, and aligning with the existing SageMaker ML lifecycle stack used by the company. The core requirement specifically targets bias drift detection, a use case that is natively supported by purpose-built SageMaker tools. The correct solution leverages a serverless trigger to run the dedicated SageMaker bias detection tool, which integrates seamlessly with deployed endpoints, S3 data stores, and existing SageMaker governance tools including the model registry, meeting both the functional and security requirements. Option Analysis: A. Correct. SageMaker Clarify is the purpose-built AWS service for ML bias and fairness detection, including native support for bias drift monitoring on deployed real-time endpoints. AWS Lambda is a serverless, event-driven service that can be invoked directly on demand from the company's web application to trigger a SageMaker Clarify job. Clarify jobs can be configured to run in isolated VPCs, access S3 training data securely without public exposure, and integrate with existing SageMaker model registry and monitoring features, fully addressing all requirements in the scenario. B. Incorrect. The sagemaker-model-monitor-analyzer built-in SageMaker image is designed for generic data quality and model quality monitoring, not bias drift detection. Bias drift monitoring requires dedicated SageMaker Clarify containers, so this option uses the wrong tool for the specified use case and will not meet the bias monitoring requirement. C. Incorrect. AWS Glue Data Quality is a service designed to monitor the quality of static datasets stored in data lakes, warehouses, and databases. It has no native integration with deployed SageMaker model endpoints or built-in functionality to detect bias drift in model predictions, so it does not address the core requirement. D. Incorrect. SageMaker notebooks are interactive development environments intended for manual experimentation and analysis, not automated on-demand workflows. A notebook-based bias comparison process is not scalable, requires manual intervention, and does not meet the requirement for an application-triggered on-demand workflow. Key Concepts: 1. SageMaker Clarify Capabilities: SageMaker Clarify is AWS's purpose-built tool for ML fairness, bias detection, and model explainability. It supports pre-training bias checks on training datasets, post-training bias checks on model outputs, and ongoing bias drift monitoring for deployed real-time and batch endpoints, making it the only native AWS tool for the bias drift use case in this question. 2. SageMaker Model Monitoring Categories: SageMaker model monitoring includes four distinct, purpose-built categories: data quality monitoring, model quality monitoring, model bias monitoring, and model explainability monitoring. Each category uses dedicated containers and tools, so selecting the correct monitoring type for the use case is critical to meet requirements. 3. Serverless SageMaker Workflow Triggers: AWS Lambda is the standard serverless solution for programmatically triggering SageMaker jobs (including Clarify, processing, training, and transform jobs) on demand, allowing seamless integration with external applications without managing underlying compute infrastructure. References: Monitor Bias Drift for Models in Production, https://docs.aws.amazon.com/sagemaker/latest/dg/clarify-model-monitor-bias-drift.html Use AWS Lambda with Amazon SageMaker, https://docs.aws.amazon.com/sagemaker/latest/dg/use-cases-lambda.html

" target="_blank" rel="nofollow noopener">https://img.examtopics.com/aws-certified-machine-learning-engineer-associate-mla-c01/image2.png">