Fundamentals of AI and ML practice questions

ml.c5.xlarge

Why wrong: Incorrect. ml.c5.xlarge does not have a GPU, so it cannot meet the GPU acceleration requirement.

ml.p3.2xlarge

Why wrong: Incorrect. While ml.p3.2xlarge has a GPU (V100), it is more expensive than ml.g5.xlarge and offers more compute than typically needed for inference, leading to higher cost without benefit.

ml.p4d.24xlarge

Why wrong: Incorrect. ml.p4d.24xlarge is a very large GPU instance with four A100 GPUs, which is excessively expensive and over-provisioned for a single-model inference endpoint.

The training data is stored in the wrong S3 bucket

Why wrong: S3 permissions or path issues do not cause container start errors.

The container image does not have the correct entry point

Why wrong: Missing entry point would cause a different error.

The GPU drivers are outdated

Why wrong: Outdated GPU drivers would cause a different error, but SageMaker manages drivers.

SageMaker Training Compiler

Why wrong: Compiler accelerates training, does not handle early stopping.

SageMaker Debugger

Why wrong: Debugger monitors training and can stop jobs based on rules, but early stopping is more directly supported by built-in algorithms.

SageMaker Experiments

Why wrong: Experiments track iterations, but do not stop jobs.

The file name contains spaces

Why wrong: File names with spaces do not cause Access Denied; they might cause a different error if not properly URL-encoded.

The policy does not grant 's3:ListBucket' permission

Why wrong: s3:ListBucket is not required to read a specific object; only s3:GetObject is needed for direct access.

The policy allows 's3:PutObject' which is not needed

Why wrong: Allowing s3:PutObject does not prevent read operations; the absence of s3:GetObject would cause the error, but this option does not state that.

The number of epochs is insufficient

Why wrong: Insufficient epochs typically result in training loss still decreasing for both training and validation sets, not the pattern of training loss decreasing while validation loss increases.

The dataset is too small

Why wrong: A small dataset may contribute to overfitting, but the direct symptom shown in the loss curves is overfitting itself, not the root cause of dataset size.

The learning rate is too low

Why wrong: A low learning rate usually causes slow convergence with both losses decreasing slowly, not the diverging pattern of training loss decreasing and validation loss increasing.

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

Why wrong: Removing the attribute may not eliminate bias because other features can act as proxies.

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

Why wrong: Collecting more data is time-consuming and costly; the team should first analyze existing data for bias.

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

Why wrong: Post-hoc adjustments may violate regulatory requirements and are less effective than addressing bias during training.

Add a feature for the marketing campaign and continue using the old model.

Why wrong: The old model was not trained with this feature, so it won't be used effectively without retraining.

Switch to a different model type, such as ARIMA, without retraining.

Why wrong: ARIMA also requires retraining on recent data to capture the new pattern.

Multiply the model's predictions by 1.2 to account for the marketing campaign.

Why wrong: This is a static adjustment that does not adapt to future changes and may introduce errors.

To tune hyperparameters

Why wrong: Hyperparameter tuning uses the validation set, not the test set.

To balance class distributions

Why wrong: Class balancing is a separate preprocessing step, not a purpose of splitting data.

To prevent overfitting during training

Why wrong: While splitting helps, the main purpose of the test set is final evaluation, not direct prevention of overfitting.

K-Means

Why wrong: K-Means is an unsupervised clustering algorithm, not suitable for regression tasks.

BlazingText

Why wrong: BlazingText is an NLP-focused algorithm for text classification and word embeddings, not general regression.

Linear Learner

Why wrong: This option duplicates option B. Although the text is identical, it is not the intended correct answer and should not be selected.

Remove outliers from training data

Why wrong: Removing outliers can help if they are erroneous, but may not guarantee improvement and could lose information.

Apply feature scaling

Why wrong: Feature scaling is important for gradient-based algorithms but not a general fix for high RMSE.

Use a different evaluation metric

Why wrong: Changing the metric does not change the model's predictions.

Label encoding

Why wrong: Label encoding assigns arbitrary integers, implying an ordinal relationship that may not exist.

Ordinal encoding

Why wrong: Ordinal encoding also imposes order, which is not appropriate for nominal data like zip codes.

One-hot encoding

Why wrong: One-hot encoding would create 50,000 binary columns, leading to high dimensionality and sparsity.

The s3:GetObject action is restricted to a specific bucket but the bucket ARN is incorrect

Why wrong: The ARN arn:aws:s3:::my-bucket/* is correctly formed for object access.

The sagemaker:CreateTrainingJob action is not allowed for the specific instance type

Why wrong: The policy allows CreateTrainingJob on all resources, so it's not the issue.

The IAM role does not have permission to write to the bucket

Why wrong: The error is about reading data, not writing.

Use Amazon SageMaker Linear Learner algorithm

Why wrong: Amazon SageMaker Linear Learner is a supervised algorithm used for regression or classification. It cannot be used effectively for unlabeled anomaly detection and does not perform dimensionality reduction.

Use Amazon SageMaker Image Classification algorithm

Why wrong: Amazon SageMaker Image Classification is a supervised algorithm for classifying images. It is not suitable for time-series anomaly detection and requires labeled data.

Use Amazon SageMaker Object Detection algorithm

Why wrong: Amazon SageMaker Object Detection is a supervised algorithm for detecting objects in images. It does not apply to unlabeled time-series anomaly detection.

K-means clustering.

Why wrong: K-means is an unsupervised learning algorithm and cannot be used to predict churn with labeled data.

Linear regression.

Why wrong: Linear regression is used for regression tasks, not binary classification problems like churn prediction.

Deep neural network with multiple hidden layers.

Why wrong: Deep neural networks are not easily interpretable and may require large amounts of data.

Reduce the VolumeSizeInGB to save costs.

Why wrong: Reducing volume size does not affect model performance; it only reduces storage cost.

Use a distributed training strategy by increasing InstanceCount to 4.

Why wrong: Distributed training speeds up training but does not directly improve model accuracy.

Switch to a more powerful instance type to reduce training time.

Why wrong: Switching instance type reduces training time but does not directly improve model metrics.

Encapsulate the custom model in a Docker container, push it to Amazon ECR, and create a custom machine learning resource in Amazon Lex to invoke the container directly.

Why wrong: Amazon Lex does not support direct invocation of custom Docker containers for machine learning models; integration is typically done through Lambda functions calling SageMaker endpoints.

Train a custom model in SageMaker using a built-in algorithm like BlazingText, then deploy it to a SageMaker endpoint and integrate with Lex via a AWS Lambda function that calls the endpoint.

Why wrong: This approach requires manual algorithm selection and tuning, increasing operational overhead for a team with limited experience.

Use Amazon Comprehend to perform sentiment analysis and entity recognition, then map the results to Lex intents using Lambda.

Why wrong: Amazon Comprehend provides pre-trained NLP capabilities, but cannot be used to train a custom NLU model tailored to the company's specific chatbot intents and slots.

Amazon SageMaker Studio Lab

Why wrong: Studio Lab provides a free notebook environment but does not automate model building.

Amazon SageMaker Clarify

Why wrong: Clarify is for bias detection and explainability, not automated model building.

Amazon SageMaker JumpStart

Why wrong: JumpStart provides prebuilt models and solutions but does not automate the full pipeline.

Amazon EFS

Why wrong: EFS is a file system, but S3 is more cost-effective for large unstructured data.

Amazon RDS

Why wrong: RDS is a relational database, not suitable for unstructured data.

Amazon EBS

Why wrong: EBS provides block storage for single instances, not ideal for large-scale data sharing.

AWS Glue

Why wrong: Glue is for ETL, not model training or deployment.

Amazon QuickSight

Why wrong: QuickSight is for visualization, not ML.

Amazon Kinesis Data Analytics

Why wrong: Kinesis Data Analytics can run SQL or Flink, but not custom ML models directly.

Increase the number of epochs

Why wrong: If loss is not decreasing, more epochs won't help.

Increase the learning rate

Why wrong: Increasing learning rate may cause divergence.

Decrease the batch size

Why wrong: Decreasing batch size can introduce noise but may not directly address plateauing loss.