All about the AWS Machine Learning Engineer - Associate exam

David Blocher is a 12-times AWS certified course author here at Pluralsight. After passing the new Machine Learning Engineer exam on the first try, here are his tips on what to expect going into the MLA-C01, and how you can ace it as well.

AWS now has a mid-level ML certification as of late August 2024! It’s called the AWS Machine Learning Engineer - Associate, and it’s currently in beta. This new certification was released alongside the new AWS Certified AI Practitioner certification. 

So, why did AWS add these certifications at all? Because there’s a growing need for engineers and technical leaders who have the skills and knowledge to implement, secure, and monitor responsible ML and AI workloads.

Who is the AWS Machine Learning Engineer - Associate for?

The new Machine Learning Engineer - Associate certification is designed for cloud engineers who work with Amazon SageMaker, and want to certify their skills in cloud architecture, data engineering, DevOps, and data science as it relates to machine learning on AWS. It’s also perfect for machine learning engineers who want to learn more about maintaining ML ecosystems on AWS.

Does this make the existing AWS Machine Learning - Speciality obsolete?

No. AWS’s famously difficult Machine Learning - Speciality certification (which has been around for six years now) still has its place. The Specialty certification focuses more on designing and running ML workloads, and has a greater emphasis on ML algorithms, hyperparameter tuning, and model design and training. 

How is the MLA-C01 exam structured?

This exam consists of 85 questions over 170 minutes. It includes the traditional AWS question types—multiple select and multiple choice— as well as three new question types called ordering, matching, and case study. If you’re thinking of taking this exam, you should definitely familiarize yourself with these new question formats.

Overall, I really liked the new question types, and the case study questions help cut down on reading time by presenting a scenario and then asking 4 or 5 questions relating to the scenario.

Just like other AWS certification exams, you can take this exam at a testing center, or take an online proctored exam from home. Personally, I prefer testing centers so I don't have to worry about technical issues (or clearing off my desk), but it's all up to your personal preference and local availability.

What the AWS Machine Learning Engineer - Associate exam covers

The exam consists of four domains: 

• Data Preparation for Machine Learning
• ML Model Development
• Deployment and Orchestration of ML Workflows
• ML Solution Monitoring, Maintenance, and Security

For a full breakdown of these domains, I highly recommend checking out the AWS exam guide.

Key things to expect when preparing to take the MLA-C01

Study up on SageMaker

An alternate name for this certification could have been the Amazon SageMaker certification. The vast majority of questions on the exam involve SageMaker, which has evolved from a single Machine Learning service to an integrated family of services designed to help you build, train, and deploy machine learning models. SageMaker provides tools for data pipelines, and infrastructure management for training, debugging, hosting, and monitoring machine learning models. 

You'll need to understand the machine learning lifecycle, from data ingestion, data storage, data preparation, model training, tuning, deployment, monitoring, and retraining, and you'll have to do this all while keeping your data secure along the way.

SageMaker Data Wrangler

You'll see a lot of SageMaker Data Wrangler, which is used for data preparation and feature engineering with minimal code writing. You'll need to understand when it's more appropriate to use Data Wrangler as opposed to other data preparation tools such as AWS Glue.

SageMaker Model Registry

You'll also need to know how to sort and manage your various machine learning models using the SageMaker Model Registry. You'll need to know the difference between Model groups and collections.

SageMaker Inference

When it comes to deploying your models, the exam heavily emphasizes using SageMaker inference, which will manage your infrastructure behind endpoints. You'll need to know when it's appropriate to use real-time, serverless, asynchronous, or batch endpoints for your models deployed to SageMaker Inference.

Beyond SageMaker, study your general AI/ML concepts and services

Outside of SageMaker, you'll need to be familiar with general machine learning concepts. You'll need to understand the difference between regression and classification models, and the metrics you would use to tune and monitor those models. You'll be tested on overfitting and underfitting, and the different techniques you can use to address these common problems.

You'll also see several questions relating to stand-alone managed AI and Machine Learning services on AWS. You'll need to know the basics of Amazon Bedrock, which is their fully-managed Generative AI service, including how to fine-tune pre-built models with proprietary data. You'll also see several questions involving Amazon Comprehend, which can be used for sentiment analysis, as well as redacting sensitive data in natural language documents.

Conclusion

This is only a snapshot of the most common services, but the breadth of this exam is far beyond what I've mentioned here. Hands-on experience will be a must for achieving this new, and highly sought-after Machine Learning certification.

Good luck on your learning and certification journey, and keep being awesome!