AWS Certified Machine Learning Engineer - Associate (MLA-C01)

The AWS Certified Machine Learning Engineer - Associate (MLA-C01) is designed for professionals who build and operationalize machine learning models on AWS. It covers the entire ML lifecycle, including data preparation, model training, deployment, and monitoring. Professionals with the symbol AWS_MLA demonstrate the technical skills needed to scale ML solutions within a cloud environment.

---------- Question 1

Your team needs to prepare a dataset for training a computer vision model that includes images of various products. The images are stored in Amazon S3, and the metadata (product name, category, etc.) is in a separate CSV file also in S3. The training data needs to be organized such that each image is associated with its corresponding metadata. Which AWS service is best suited to efficiently handle this data preparation task?

1. Amazon EMR
2. AWS Glue DataBrew
3. Amazon Athena
4. Amazon SageMaker Data Wrangler

---------- Question 2

You are developing a recommendation system using a collaborative filtering approach. After training several models, you observe that one model consistently outperforms others on metrics such as precision and recall, but has significantly higher inference latency. Considering this trade-off, what strategy would you employ to deploy this higher-performing but slower model?

1. Discard the higher-performing model and use the fastest model regardless of accuracy.
2. Deploy the model to a real-time endpoint using powerful GPU instances.
3. Deploy the model to a batch inference endpoint.
4. Use model compression techniques to reduce the model size without significantly impacting accuracy and then deploy to a real-time endpoint.

---------- Question 3

You need to deploy a real-time fraud detection model that requires low latency and high throughput. Which SageMaker endpoint configuration would be MOST suitable for this requirement?

1. Serverless inference endpoint with a single instance.
2. Real-time endpoint with auto-scaling enabled.
3. Batch transform job.
4. Asynchronous endpoint using SQS for queuing.

---------- Question 4

Your model training dataset shows a significant class imbalance where fraudulent transactions represent only 1% of the total data. To address this imbalance and improve model performance on the minority class, what strategy should you prioritize?

1. Ignore the imbalance; the model will adapt.
2. Remove the majority class samples.
3. Employ oversampling techniques like SMOTE (Synthetic Minority Over-sampling Technique).
4. Use cost-sensitive learning to assign higher weights to the minority class during training.

---------- Question 5

You've trained three different models for a customer churn prediction problem: a logistic regression model, a random forest model, and a gradient boosting machine (GBM) model. All three models have similar accuracy, but the GBM model has higher recall and precision for the minority class (customers likely to churn). Considering business priorities (minimizing churn), which model should you choose for deployment and why?

1. The logistic regression model, as it is simpler and easier to interpret.
2. The random forest model, as it offers a good balance between accuracy, interpretability, and performance.
3. The GBM model, because its higher recall and precision for the minority class better aligns with the business objective of minimizing churn.
4. All three models are equally suitable given the comparable accuracy scores.

---------- Question 6

Your team is deploying a new version of a machine learning model. To minimize risk, you want to gradually roll out the new model to a subset of users before deploying it to the entire user base. Which deployment strategy best suits this requirement?

1. Linear Deployment
2. Blue/Green Deployment
3. Canary Deployment
4. Rolling Back Deployment

---------- Question 7

Your team is developing a customer churn prediction model. During feature engineering, you discover several highly correlated features. What's the BEST strategy to handle this correlation to improve model performance and avoid overfitting?

1. Keep all correlated features; the model will automatically handle the redundancy.
2. Remove all features that show any degree of correlation.
3. Perform Principal Component Analysis (PCA) to reduce dimensionality and retain the most important variance.
4. Use one-hot encoding on all correlated features to address the issue.

---------- Question 8

You're developing a natural language processing (NLP) model to analyze customer reviews. The model needs to handle various sentiment expressions and handle a large vocabulary. You have a large dataset and are concerned about training time. Which combination of approaches would be MOST effective in reducing training time while maintaining acceptable model accuracy?

1. Use a simple linear regression model and train it on a subset of your data.
2. Use a complex recurrent neural network (RNN) model trained with a single GPU.
3. Employ distributed training with SageMaker on a cluster of multiple GPUs using a pre-trained Transformer model fine-tuned on your dataset.
4. Use a Random Forest model with default hyperparameters.

---------- Question 9

You're building a fraud detection model using a large, imbalanced dataset. The positive class (fraudulent transactions) represents only 0.1% of the data. Simply training a model on this dataset will lead to poor performance. Which strategy would be MOST effective in addressing this class imbalance and improving model accuracy?

1. Over-sample the majority class (non-fraudulent transactions).
2. Under-sample the minority class (fraudulent transactions).
3. Use a cost-sensitive learning algorithm and SMOTE (Synthetic Minority Over-sampling Technique).
4. Ignore the class imbalance; the model will learn to identify fraud regardless.

---------- Question 10

Your company is migrating a large on-premises data warehouse (Terabytes of data) to AWS for machine learning. The data is structured and needs to be readily accessible for multiple SageMaker training jobs. Cost is a major concern, and you need to minimize data transfer costs during the migration. Which combination of AWS services best addresses these requirements, considering both initial migration and ongoing access?

1. Use Amazon S3 Standard for storage and Amazon S3 Transfer Acceleration for migration. Access the data directly from S3 for training.
2. Use Amazon S3 Glacier Deep Archive for cost-effective long-term storage and retrieve data on demand for training. Use Amazon EMR for processing.
3. Migrate the data to Amazon Redshift for fast query processing, then export data subsets to S3 for SageMaker training.
4. Use Amazon FSx for NetApp ONTAP for high-performance storage and migrate the data using AWS DataSync. Access the data directly from FSx for training.

---

Are they useful?

Click here to get 360 more questions to pass this certification at the first try! Explanation for each answer is included!

Follow the below LINKEDIN channel to stay updated about 89+ exams!

Digital Intelligence Network