AIF-C01 Hard Questions — 25 Difficult Questions

Question 1

A company is deploying a machine learning model for real-time fraud detection. The model must make predictions with latency under 10 milliseconds. The data scientist trained a gradient boosting model that achieves high accuracy but has inference latency of 50 milliseconds. The team has access to a larger instance type with more CPU cores. Which approach should the data scientist take to reduce inference latency while maintaining accuracy?

Accepted Answer

Deploy the model on a larger instance with more CPU cores.. Option B is correct because increasing the number of CPU cores allows the gradient boosting model to parallelize tree evaluation across multiple cores, reducing inference latency. Since the model is already trained and accurate, this hardware scaling directly addresses the 50 ms bottleneck without altering the model's structure or accuracy.

Answer

Switch to batch inference and run predictions every 100 milliseconds.

Answer

Reduce the maximum tree depth and retrain the model.

Answer

Apply post-training pruning to remove redundant trees.

Question 2

An organization wants to use Amazon Rekognition to analyze images of people for a security application. They must comply with GDPR. What is the best practice?

Accepted Answer

Ensure all images are anonymized before analysis. Option C is correct because GDPR requires that personal data, including facial images, be processed lawfully and with appropriate safeguards. Anonymizing images before analysis with Amazon Rekognition ensures that the data cannot be linked back to an identifiable person, thereby reducing GDPR compliance risk. This aligns with the principle of data minimization and privacy by design.

Answer

Store images indefinitely for audit

Answer

Use celebrity recognition

Answer

Use face detection only

Question 3

A financial services company is deploying a machine learning model that must comply with SOC 2 and PCI DSS. They need to ensure that the model artifacts and training data are encrypted, access is audited, and the environment is protected from network threats. Which THREE AWS services should they implement?

Accepted Answer

Amazon GuardDuty. Amazon GuardDuty is a threat detection service that continuously monitors for malicious activity and unauthorized behavior, which helps protect the environment from network threats as required by SOC 2 and PCI DSS. By analyzing VPC Flow Logs, DNS logs, and CloudTrail events, it can detect anomalies such as port scanning or data exfiltration, directly addressing the need for network threat protection in a compliant ML deployment.

Answer

Amazon DynamoDB Accelerator (DAX)

Answer

AWS WAF

Question 4

A security engineer creates the above IAM policy to allow a user to invoke an Amazon Bedrock model. However, invocation fails. What is the issue?

Accepted Answer

The ARN should use "foundation-model" instead of "model".. Option C is correct because the IAM policy's resource ARN incorrectly uses 'model' in the path, but Amazon Bedrock requires 'foundation-model' to reference foundation models. The correct ARN format for invoking a Bedrock foundation model is 'arn:aws:bedrock:region::foundation-model/model-id'. Using 'model' instead of 'foundation-model' causes the policy to not match any valid Bedrock resource, resulting in an invocation failure.

Answer

The action should be "bedrock:InvokeModelWithResponseStream".

Answer

The resource ARN is missing the account ID.

Answer

The statement is missing a condition for the model ID.

Question 5

A research team is using Amazon Bedrock to analyze scientific papers. They want the model to generate answers based only on papers published after 2023. Which approach should they use?

Accepted Answer

Use Amazon Bedrock Knowledge Bases with a metadata filter to retrieve only papers published after 2023, and generate responses based on retrieved content.. Option D is correct because Amazon Bedrock Knowledge Bases with a metadata filter allows you to restrict retrieval to only documents that match specific metadata criteria, such as publication year. By filtering the vector search to only include papers published after 2023, the model generates responses based solely on that retrieved content, ensuring it does not rely on pre-2023 data. This approach is the only one that guarantees the model's answers are grounded exclusively in the specified time range.

Answer

Fine-tune the model on a dataset of post-2023 papers and deploy it.

Answer

Set the maxTokens to a low value to force the model to rely on recent context.

Answer

Include a system prompt instructing the model to ignore data before 2023.

Question 6

A data scientist is fine-tuning a foundation model on a custom dataset using Amazon SageMaker. After training, the model shows high accuracy on training data but poor on validation. Which action should be taken?

Accepted Answer

Reduce training epochs or add regularization. The model is overfitting, as indicated by high training accuracy but poor validation performance. Reducing training epochs or adding regularization (e.g., L1/L2 weight decay) directly addresses overfitting by limiting the model's capacity to memorize noise. In Amazon SageMaker, this can be implemented via hyperparameter tuning or by modifying the training script to include regularization terms.

Answer

Add dropout layers

Answer

Increase learning rate

Answer

Use a different foundation model

Question 7

A team is fine-tuning a foundation model using SageMaker. They want to minimize training time while keeping the model's original knowledge. Which technique is BEST suited?

Accepted Answer

Use Parameter Efficient Fine-Tuning (PEFT) such as LoRA. Parameter Efficient Fine-Tuning (PEFT) methods like LoRA (Low-Rank Adaptation) are best suited because they freeze the pre-trained model weights and inject trainable low-rank matrices into specific layers, drastically reducing the number of trainable parameters. This minimizes training time and computational cost while preserving the model's original knowledge, as only a small fraction of parameters are updated during fine-tuning.

Answer

Use distributed training across multiple GPUs

Answer

Use prompt engineering instead of fine-tuning

Answer

Full fine-tuning on the new dataset

Question 8

Which TWO of the following are valid methods to reduce the risk of foundation models generating harmful or biased content?

Accepted Answer

Use a content filter. Option B is correct because content filters act as a safety layer that intercepts and blocks harmful or biased outputs before they reach the user. These filters can be rule-based or use a separate classifier model trained to detect toxic, hateful, or biased language, reducing the risk of harmful content generation without altering the underlying model.

Answer

Use a smaller model

Answer

Fine-tune the model on a biased dataset

Answer

Disable all logging

Question 9

Refer to the exhibit. A developer deploys this CloudFormation stack but the agent fails to query the knowledge base. What is a likely cause?

Accepted Answer

The agent role does not have permissions to invoke the knowledge base. The correct answer is B because the agent role must have an IAM policy that grants the `bedrock:Retrieve` and `bedrock:RetrieveAndGenerate` permissions on the knowledge base. Without these permissions, the agent cannot invoke the knowledge base, even if the KnowledgeBaseId is correctly passed and the embedding model is available.

Answer

The KnowledgeBaseId is not passed correctly

Answer

The embedding model is not available in the region

Answer

The OpenSearch collection type should be SEARCH not VECTORSEARCH

Question 10

A data science team is fine-tuning a foundation model on Amazon SageMaker. Which THREE steps are part of the best practice? (Choose three.)

Accepted Answer

Monitor for catastrophic forgetting during fine-tuning.. Option B is correct because catastrophic forgetting is a known risk when fine-tuning foundation models, where the model loses previously learned knowledge while adapting to new data. Monitoring for this during fine-tuning on SageMaker allows the team to detect performance degradation on the original task and adjust training accordingly, ensuring the model retains its general capabilities.

Answer

Increase model size to improve performance.

Answer

Deploy the model to production immediately after fine-tuning.

Question 11

A company is building a multi-modal application that processes images and text to answer questions about product defects. Which foundation model approach is BEST?

Accepted Answer

Use a multi-modal foundation model that processes both images and text. Option C is correct because multi-modal foundation models (e.g., CLIP, Flamingo, GPT-4V) are specifically designed to jointly process and reason over images and text in a unified architecture. This allows the model to directly correlate visual defects with textual descriptions without intermediate lossy transformations, making it the most effective approach for a multi-modal QA task.

Answer

Use an image captioning model and then analyze the caption text

Answer

Use a text-to-image generation model and analyze the generated image

Answer

Use a separate image analysis model and a text model, then combine outputs

Question 12

Refer to the exhibit. You are trying to invoke a foundation model via Amazon Bedrock but receive this error. What should you do to resolve it?

Accepted Answer

Request model access in the Bedrock console. The error indicates that the user has not been granted access to the specific foundation model in Amazon Bedrock. Even with valid IAM permissions, each model requires explicit access approval via the Bedrock console's 'Model access' section. Option B is correct because requesting model access there provisions the necessary service-level authorization.

Answer

Increase the service quota for Bedrock

Answer

Attach the AmazonBedrockFullAccess IAM policy

Answer

Use a different AWS Region

Question 13

A company is deploying a generative AI model on Amazon Bedrock. The model is accessed by an application running on Amazon ECS Fargate. The security team requires that all data sent to and from the Bedrock model be encrypted in transit, and that the ECS task does not have internet access. The ECS tasks are launched in a private VPC subnet with a VPC endpoint for Bedrock configured. However, when the application attempts to invoke the Bedrock model, the call fails with a timeout error. The VPC endpoint policy is set to allow all actions from all principals. What is the most likely cause of the failure?

Accepted Answer

The security group associated with the VPC endpoint does not allow inbound HTTPS traffic from the ECS tasks. The timeout error indicates a network connectivity issue, not an authentication or authorization problem. Since the ECS tasks are in a private subnet without internet access and a VPC endpoint for Bedrock is configured, the most likely cause is that the security group associated with the VPC endpoint does not allow inbound HTTPS (port 443) traffic from the ECS tasks. Without this inbound rule, the TCP handshake fails, resulting in a timeout.

Answer

The ECS tasks are not configured to use a NAT gateway to reach the Bedrock endpoint

Answer

The VPC endpoint policy does not grant the ECS task IAM role permission to invoke Bedrock

Answer

The ECS task IAM role does not have the bedrock:InvokeModel permission

Question 14

A company is training a deep learning model on Amazon SageMaker using a custom Docker container. The training job fails with the error 'CannotStartContainerError: API error (500): failed to create shim task'. The team verifies that the container image is compatible with the selected instance type. What is the most likely cause of this error?

Accepted Answer

The instance type does not have enough memory for the container. The error 'CannotStartContainerError: API error (500): failed to create shim task' typically occurs when the Docker container cannot be initialized due to resource constraints, most commonly insufficient memory on the selected instance type. Even if the container image is compatible with the instance, the container's memory request may exceed the available memory, causing the container runtime (containerd) to fail when creating the shim task. This is a known issue in SageMaker when the training job's resource requirements are not aligned with the instance's capacity.

Answer

The training data is stored in the wrong S3 bucket

Answer

The container image does not have the correct entry point

Answer

The GPU drivers are outdated

Question 15

A financial services company uses a machine learning model to approve loan applications. The model is a gradient boosting classifier trained on historical loan data. Recently, the company noticed that the model's approval rate for applicants from a certain demographic group is significantly lower than for other groups, even though the model's overall accuracy remains high. The data science team has been asked to address this potential bias while minimizing the impact on overall model performance. The team has access to the training data and the trained model. They have limited time and budget. Which course of action should the team take first?

Accepted Answer

Analyze the training data for bias and retrain the model using bias mitigation techniques such as reweighting.. The most efficient first step is to analyze the training data for bias and then retrain the model with bias mitigation techniques like reweighting. Option A is wrong because collecting more data is resource-intensive and may not address bias. Option C is wrong because feature engineering may not help if the bias is in the labels. Option D is wrong because post-hoc adjustments can introduce other issues and may not be as effective as addressing bias during training.

Answer

Remove the sensitive attribute from the training data and retrain the model.

Answer

Collect more data from the under-represented demographic group and retrain the model.

Answer

Adjust the model's decision threshold for the affected group after deployment.

Question 16

A team is using Amazon Bedrock to generate images from text prompts. The generated images often contain artifacts and do not match the prompt description. Which combination of steps should the team take to improve image quality?

Accepted Answer

Refine the prompt with more descriptive language and adjust the CFG scale and inference steps.. Option C is correct because refining the prompt with more descriptive language helps the model better interpret the user's intent, while adjusting the CFG (Classifier-Free Guidance) scale controls how strictly the model adheres to the prompt, and increasing inference steps allows the diffusion process to produce higher-quality, artifact-free images. These are standard hyperparameters in diffusion-based image generation models on Amazon Bedrock, directly addressing both artifacts and prompt mismatch.

Answer

Fine-tune the model using SageMaker Ground Truth and increase the training epochs.

Answer

Increase the max token count and use a larger model variant.

Answer

Use a different foundation model and increase the image resolution.

Question 17

A company uses Amazon Bedrock to generate product descriptions. They want to ensure the outputs consistently follow a specific brand tone (professional yet friendly). They have a small set of example descriptions (few-shot examples) but do not want to fine-tune the model. Which strategy best achieves consistent tone without modifying the base model?

Accepted Answer

Use a system prompt that defines the brand tone and include few-shot examples in the prompt.. Using a system prompt and few-shot examples in the prompt template (Option D) provides explicit guidance to the model at inference time, shaping the tone without any model updates. Option A (retrieval-augmented generation) is for incorporating external knowledge, not tone. Option B (prompt chaining) adds complexity and may not directly enforce tone. Option C (fine-tuning) requires modifying model weights, which is not desired.

Answer

Fine-tune the model on a dataset of product descriptions that exemplify the desired tone.

Answer

Implement prompt chaining by breaking the task into multiple steps, each with its own prompt.

Answer

Use retrieval-augmented generation (RAG) to pull example descriptions from a database and prepend them to the prompt.

Question 18

A company uses Amazon SageMaker to host a real-time inference endpoint. The model was trained on sensitive data, and the company wants to ensure that the data sent to the endpoint is encrypted in transit. Additionally, the company wants to restrict access to the endpoint to only traffic originating from a specific VPC. Which configuration meets these requirements?

Accepted Answer

Create the SageMaker endpoint in a VPC, associate a security group that allows inbound HTTPS from the VPC CIDR, and configure the endpoint to use HTTPS. Option A is correct because creating the SageMaker endpoint within a VPC and associating a security group that restricts inbound HTTPS traffic to the VPC CIDR ensures that only traffic from within that VPC can reach the endpoint. Configuring the endpoint to use HTTPS encrypts data in transit using TLS, meeting both the encryption and VPC-only access requirements.

Answer

Configure the SageMaker endpoint to use mutual TLS (mTLS) with client certificates

Answer

Place the SageMaker endpoint behind an Amazon CloudFront distribution with an origin access identity

Answer

Use AWS STS to generate temporary credentials and require the client to sign requests with them

Question 19

An ML engineer wants to store training data in a format optimized for linear data scanning and columnar access in SageMaker. Which format is most appropriate?

Accepted Answer

Parquet. Parquet is a columnar storage format optimized for both linear data scanning and columnar access, making it ideal for training data in SageMaker. It reduces I/O by storing data by columns rather than rows, enabling efficient retrieval of specific features during model training.

Answer

JSON

Answer

Image (JPEG/PNG)

Answer

CSV

Question 20

A data engineer is using Amazon SageMaker Data Wrangler to prepare tabular data for ML. Which THREE data transformations are natively supported? (Choose three.)

Accepted Answer

One-hot encoding for categorical features. Option A is correct because Amazon SageMaker Data Wrangler includes built-in support for one-hot encoding as a native transformation for categorical features. This transformation automatically creates binary columns for each category, which is essential for preparing tabular data for machine learning models that require numerical input.

Answer

Audio feature extraction

Answer

Image resizing and normalization