单选题 A publishing company uses a text-to-text foundation model (FM) on Amazon Bedrock for summarization, but the model misinterprets casual language, local expressions, and abbreviations in customer feedback, leading to inaccurate summaries.
Which solution provides the most efficient and cost-effective approach to improve the model’s understanding of customer feedback?
A、 Launch a new large-scale training job in Amazon SageMaker AI using the model-parallelism library to build a domain-specific language model trained entirely on historical customer reviews.
B、 Use Amazon SageMaker Data Wrangler to preprocess customer feedback data, remove slang and abbreviations, and standardize the language before sending it to the Bedrock model for summarization.
C、 Implement Custom Entity Recognition (CER) to extract slang terms and abbreviations from customer feedback and use these extracted entities as metadata inputs to Bedrock during text generation.
D、 Customize the current foundation model by applying fine-tuning using labeled datasets of customer feedback that reflect informal wording, abbreviations, and expressions.
单选题 A financial analytics firm needs to evaluate a new fraud detection model's prediction accuracy and latency in production without affecting the throughput of the currently deployed model or requiring any changes to how clients invoke the inference endpoint.
Which option satisfies the given requirements?
A、 Deploy both models in separate SageMaker endpoints and use Amazon CloudWatch metrics to compare their results in post-processing.
B、 Register the new model version in SageMaker Model Registry and configure an event trigger in AWS Lambda to automatically swap the endpoint to the new model after initial validation.
C、 Configure an Amazon API Gateway endpoint that splits traffic between the current and the new SageMaker endpoints for A/B testing.
D、 Modify the existing SageMaker AI endpoint configuration by adding the new model as a ProductionVariant through the ProductionVariant API, and set a small InitialVariantWeight compared to the existing model’s ProductionVariant VariantWeight to control the percentage of traffic routed to it.
单选题 A financial organization needs to migrate a custom, real-time fraud detection model (less than 5 GB, up to 50 concurrent requests) from on-premises infrastructure to AWS, prioritizing minimal infrastructure management.
Which solution meets the requirements with the least operational overhead?
A、 Deploy the custom fraud detection model in Amazon SageMaker Neo to optimize the model, then host the optimized model on a SageMaker real-time endpoint.
B、 Create a model configuration within Amazon SageMaker AI, then deploy the custom fraud detection model on an asynchronous SageMaker endpoint.
C、 Deploy the fraud detection model on a highly available Amazon EC2 instance in an auto-scaling group. Configure an application load balancer to route the incoming requests to the EC2 instance.
D、 Create a model configuration within Amazon SageMaker AI, then deploy the custom fraud detection model on a serverless SageMaker endpoint.
单选题 A data scientist needs to develop a fraud detection model on SageMaker with a severely imbalanced dataset (fraudulent transactions are rare). They must minimize operational overhead and ensure the model is fair and unbiased.
Which approach will fulfill the given requirements?
A、 Use SageMaker Studio to preprocess the data and apply SMOTE, then use SageMaker Reinforcement Learning to build a fraud detection model and check for bias with SageMaker Clarify.
B、 Use SageMaker Studio for data processing and model development, integrating the synthetic minority oversampling technique (SMOTE) into the workflow. Once the model is trained, use Amazon Augmented AI (Amazon A2I) for bias detection before deployment.
C、 Use SageMaker Studio to preprocess the data and apply the synthetic minority oversampling technique (SMOTE) to balance the dataset. Build the model using SageMaker Pipelines and use SageMaker Clarify for bias detection before deployment.
D、 Use SageMaker Studio to preprocess and balance the data using the synthetic minority oversampling technique (SMOTE), then develop a fraud detection model with SageMaker JumpStart. Afterward, use SageMaker Clarify to check for bias and finalize the model for deployment.
单选题 A generative AI chatbot on Amazon Bedrock experiences performance degradation and crashes due to memory-related errors when processing complex or high-volume queries simultaneously.
Which solution will resolve this issue and improve the model’s stability by managing higher memory demands during peak usage?
A、 Add more processing power to the Bedrock AgentCore agent.
B、 Increase the number of Bedrock AgentCore agent instances.
C、 Upgrade the storage capacity of the Bedrock AgentCore agent.
D、 Expand the memory capacity of the Bedrock AgentCore agent.
单选题 An AI developer needs a scalable, secure way to collect telemetry data (temperature, pressure) from devices in remote locations with unstable connectivity, store it in Amazon S3, and minimize infrastructure management.
Which solution meets the given requirements?
A、 Stream the telemetry data over Message Queuing Telemetry Transport (MQTT) to AWS IoT Core, forward it to an Amazon Kinesis Data Stream, and then configure an AWS Lambda function to process the data and send it to S3.
B、 Set up a serverless application with Amazon API Gateway to collect telemetry data from the devices, then use AWS Lambda to process and deliver the data to S3.
C、 Use AWS IoT Greengrass on each device to preprocess telemetry data locally, then batch upload the data to S3 using AWS SDK calls from the edge.
D、 Route telemetry data over Message Queuing Telemetry Transport (MQTT) to AWS IoT Core, configure a rule in IoT Core to direct the data to an Amazon Data Firehose stream that delivers data to an S3.
单选题 SageMaker notebook instances are deployed inside an isolated VPC with interface endpoints, yet unauthorized external users can still access them through the internet.
How can the team limit access to the SageMaker notebook instances, ensuring only authorized VPC users can connect?
A、 Apply VPC Endpoint Policies to control which IAM users or services can access SageMaker AI through the VPC interface endpoint, providing more granular access control for interactions with SageMaker AI.
B、 Set up VPC Traffic Mirroring to capture traffic to and from the notebook instances and identify unauthorized access attempts, enabling enhanced monitoring.
C、 Update the security group for the notebook instances to restrict incoming traffic to only the CIDR blocks associated with the VPC. Apply this security group across all interfaces linked to the SageMaker notebook instances.
D、 Configure an IAM policy that allows sagemaker:CreatePresignedNotebookInstanceUrl and sagemaker:DescribeNotebookInstance actions exclusively from VPC interface endpoints. Ensure this policy is applied to the appropriate IAM users, groups, and roles.
单选题 An ML engineer uses Amazon SageMaker Data Wrangler to explore a numerical feature (image brightness) before applying normalization, as it affects model convergence.
Which action should the engineer take to best understand the range and distribution of the brightness feature values before transformation?
A、 The engineer should use Comprehend to perform sentiment analysis on the brightness values to determine if normalization is needed.
B、 The engineer should export the dataset to Amazon S3 and use AWS Glue DataBrew to create a box plot visualization of the brightness feature.
C、 The engineer should use SageMaker Clarify to detect data bias in the brightness feature before performing any normalization.
D、 The engineer should use the SageMaker Data Wrangler histogram visualization to inspect the range of values for the brightness feature and identify any outliers.
2026-02-28
共76道
