The ideal candidate should have a minimum of one year of experience working with Amazon SageMaker and other AWS services for machine learning. They should also have at least one year of experience in a relevant role, such as backend software developer, DevOps developer, data engineer, or data scientist. Aside from that, it is recommended that you already possess general knowledge on these concepts enumerated below:

General IT Knowledge 

• Basic knowledge of popular machine learning algorithms and when to use them

• Understanding of data engineering, including common data formats, how to bring in data, and how to transform it for ML pipelines.

• Familiarity with querying and manipulating data.

• Awareness of software development best practices for writing modular, reusable code, and handling deployment and debugging.

• Experience with setting up and monitoring ML resources both in the cloud and on-premises.

• Familiarity with CI/CD pipelines, including its management, and infrastructure as code (IaC)

General AWS Knowledge

NEW Question types for the AWS Certification Exams!

Last July, the AWS announced that their Certification exams are adding three new question types which are: ordering, matching, and case study. These new question types were included to reduce your reading time while covering more key concepts. Ordering and matching questions are a more efficient method for assessing procedural understanding and pairing skills compared to multiple-choice or multiple-response questions. Meanwhile, case studies allow multiple questions to be asked based on one scenario, so you won’t need to read a new scenario for every question. These new question types will carry the same point value as multiple-choice and multiple-response questions, and they will be integrated throughout the exam alongside the existing question formats. And these new question types will first be featured on this new AWS Certified Machine Learning Engineer-Associate exam (along with the AWS Certified AI Practitioner Exam). That’s why candidates should now adjust their preparation strategies by familiarizing yourself with these new question formats, emphasizing on learning the sequences or processes related to AWS services, and improving more your critical thinking and analysis. This will validate how you can apply your knowledge to develop effective solutions to real-life scenarios and problems.

Even with the addition of new question types, there’s no need to worry as it won’t lead to considerable changes in the exams since the total number of exam questions and allotted time for taking the exam still stays the same with the other Associate exams. The MLA-C01 exam includes 65 questions and your exam results are presented as a scaled score ranging from 100 to 1,000, with a minimum passing score set at 720.

AWS Certified Machine Learning Engineer Associate MLA-C01 Exam Domains

The official exam guide for the AWS Certified Machine Learning Engineer Associate MLA-C01 provides a comprehensive list of exam domains, relevant topics, and services that require your focus. The certification exam comprises of four (4) exam domains and their respective weightings, as shown below:

Since the first domain which is “Data Preparation for Machine Learning (ML)” holds the highest exam coverage of 28%, you should give importance to the topics included in this section. However, it’s equally important to devote sufficient attention to the other domains, as they also contribute significantly to your overall understanding and performance on the exam. Each domain plays a crucial role in your preparation, and neglecting them could leave gaps in your knowledge. Listed below are the exam domains and their respective skills and knowledge that you should posses.

MLA-C01 Domain 1: Data Preparation for Machine Learning (ML)

• 1.1: Ingest and Store Data

Data cleaning and transformation techniques help improve datasets by detecting outliers, filling in missing values, combining data, and removing duplicates. Feature engineering enhances analysis through scaling, splitting, binning, and transforming data. Encoding methods like one-hot and label encoding convert categorical data into usable formats. Tools such as SageMaker Data Wrangler and AWS Glue assist in data exploration and transformation, while AWS Lambda and Spark manage real-time streaming data. Finally, data annotation services create labeled datasets for machine learning.

• 1.2: Transform data and perform feature engineering.

Understanding data processing involves knowing how to clean and transform datasets by spotting outliers, filling in missing values, and eliminating duplicates. It also requires mastering feature engineering techniques like scaling, splitting, and normalizing data, as well as encoding methods such as one-hot and label encoding to make data more usable. Familiarity with tools like SageMaker Data Wrangler and AWS Glue helps in exploring and transforming data effectively, while services like AWS Lambda and Spark are key for managing streaming data. Additionally, having skills in AWS tools for transforming data, managing features, and validating datasets, such as using SageMaker Feature Store and SageMaker Ground Truth, is important for creating high-quality labeled datasets.

• 1.3: Ensure data integrity and prepare data for modeling.

Knowledge in data processing includes understanding bias metrics for numeric, text, and image data, such as class imbalance and label differences. It also involves strategies to fix these biases using methods like synthetic data generation and resampling, as well as techniques for encrypting, classifying, anonymizing, and masking data. Being aware of compliance requirements related to personally identifiable information (PII), protected health information (PHI), and data residency is essential. Skills in this area involve validating data quality with tools like AWS Glue DataBrew, identifying and addressing biases with AWS tools like SageMaker Clarify, and preparing data to reduce prediction bias through techniques like splitting, shuffling, and augmentation, along with configuring data for model training using Amazon EFS and Amazon FSx.

MLA-C01 Domain 2: ML Model Deployment

• 2.1: Choose a modeling approach

Knowledge of machine learning (ML) includes understanding how different algorithms can address business challenges and using AWS AI services like Amazon Translate, Amazon Transcribe, Amazon Rekognition, and Amazon Bedrock for specific solutions. It also involves considering model interpretability during selection. Key skills in this area include assessing data and problem complexity to evaluate ML feasibility, choosing suitable models or algorithms, and selecting built-in algorithms and templates from SageMaker JumpStart and Amazon Bedrock. Additionally, it involves considering costs and identifying AI services that fulfill common business needs.

• 2.2: Train and refine models

Knowledge of model training encompasses key components like epochs, steps, and batch size, as well as strategies to reduce training time through early stopping and distributed training. This involves recognizing factors influencing model size and enhancing performance with regularization techniques such as dropout and weight decay. Familiarity with hyperparameter tuning methods, such as random search and Bayesian optimization, is essential, along with understanding the impact of hyperparameters on performance and how to integrate externally built models into SageMaker. Skills in this area include using SageMaker’s built-in algorithms and common ML libraries for development, employing SageMaker script mode with frameworks like TensorFlow and PyTorch, and fine-tuning pre-trained models using Amazon Bedrock or SageMaker JumpStart. Additionally, expertise in hyperparameter tuning with SageMaker’s automatic model tuning (AMT), preventing overfitting and underfitting, combining models through ensembling and boosting, reducing model size via pruning and compression, and managing model versions is important.

• 2.3: Analyze model performance

Knowledge of model evaluation involves understanding various techniques and metrics, including confusion matrices, heat maps, F1 score, accuracy, precision, recall, Root Mean Square Error (RMSE), receiver operating characteristic (ROC), and Area Under the ROC Curve (AUC). It also encompasses methods for creating performance baselines, identifying overfitting and underfitting, and utilizing metrics in SageMaker Clarify to gain insights into machine learning training data and models, as well as recognizing convergence issues. Skills in this area include selecting and interpreting evaluation metrics, detecting model bias, and weighing trade-offs between model performance, training time, and cost. Additionally, it involves conducting reproducible experiments with AWS services, comparing the performance of a shadow variant to a production variant, interpreting model outputs using SageMaker Clarify, and debugging convergence issues with SageMaker Model Debugger.