Ngân hàng đề — AWS Certified Machine Learning Engineer Associate

Amazon SageMaker for handwriting recognition, Amazon Comprehend for sentiment analysis, and Amazon Kendra for querying the extracted metadata.

Amazon Rekognition for detecting text and images, Amazon Comprehend for analyzing the extracted text, and Amazon RDS for storing structured data.

Amazon Textract for extracting both printed and handwritten text, Amazon Comprehend for entity recognition, and Amazon S3 for storing the extracted metadata.

Amazon Comprehend for entity extraction, Amazon Textract for analyzing images and text, and Amazon Elasticsearch for indexing and querying the results.

Use Kinesis Firehose to scale automatically and reduce latency for each consumer.

Enable enhanced fan-out for each consumer to achieve dedicated read throughput and lower latency.

Use AWS Glue for ETL and real-time data analysis of the streaming data.

Use standard Kinesis consumers and scale the number of shards as needed.

Use a custom SageMaker algorithm to filter out low-quality data before making predictions in the production environment.

Enable SageMaker Feature Store to log and validate all incoming data used for predictions and correct data quality issues in real-time.

Use SageMaker Model Monitor to evaluate incoming data for missing values, outliers, and deviations from the baseline training data.

Set up Amazon CloudWatch to automatically retrain the model whenever the incoming data has quality issues.

Add the new features to the dataset, retrain the model, and compare the new model’s performance.

Enable Amazon SageMaker’s AutoPilot feature to automatically detect and add holiday and competitor pricing features.

Use Amazon Forecast, which automatically incorporates external variables, such as holidays, into the forecasting model.

Use Amazon SageMaker’s hyperparameter tuning to adjust the model based on new features.

Grid Search

Bayesian Optimization

Random Search

Hyperband

Custom algorithms provide faster training times and more efficient resource usage compared to built-in algorithms.

Custom frameworks such as PyTorch and TensorFlow must be used with SageMaker to enable hyperparameter tuning.

Built-in algorithms can only be used for simple tasks like linear regression and are not suitable for complex machine learning tasks.

Pre-configured environments and infrastructure management are handled by AWS, eliminating the need for custom containers.

Built-in algorithms like XGBoost and Linear Learner abstract away infrastructure setup and allow quick model deployment.

Use Amazon Polly to convert speech to text, AWS Glue to manage translations, and Amazon Transcribe to generate speech responses.

Use AWS Lambda to convert speech to text, Amazon Translate to handle text translations, and Amazon SageMaker to generate speech responses.

Use Amazon Transcribe to convert speech to text, Amazon Translate to translate the text, and Amazon Polly to convert the translated text to speech.

Use Amazon Transcribe to convert speech to text, Amazon Lex to handle translations, and Amazon Polly to generate the voice responses.

Switch to a different model algorithm that automatically reduces the importance of "annual income" in predictions.

Fine-tune the hyperparameters of the existing model to lower the weight of the "annual income" feature.

Retrain the model with a larger dataset that excludes the "annual income" feature to prevent bias in predictions.

Use SageMaker Model Monitor to alert the team about this shift and retrain the model with updated data to ensure feature importance remains balanced.

Use SageMaker Data Wrangler for preprocessing, Amazon Redshift for feature storage, and SageMaker Notebooks to train and deploy the machine learning model.

Use SageMaker Notebooks for data preprocessing, Amazon DynamoDB for feature storage, and SageMaker Feature Store to train the machine learning model.

Use SageMaker Data Wrangler for data preprocessing, SageMaker Feature Store to store features, and SageMaker Notebooks to train and deploy the machine learning model.

Use Amazon QuickSight to preprocess and clean the data, SageMaker Feature Store for feature storage, and SageMaker Autopilot to train the machine learning model.

Dataset, algorithm, performance metric

Algorithm, hyperparameter range, performance metric

Hyperparameter range, dataset, algorithm

Algorithm, performance metric, model parameters