ML infrastructure @ AWS Summit 2021

Rustem Feyzkhanov
2 min readMay 6, 2021

Organizing infrastructure for ML applications is a challenging task. Depending on your pipeline you may need to combine different types of virtual machines, organize research infrastructure as well as monitor the system. You may also need to have a way of maintaining ML infrastructure so it’s to deploy updates. Here are the top 5talks from AWS Summit which would help you to stay in the loop regarding current trends.

AWS Summit is free and sessions can be watched on-demand. Feel free to register at the link.

AIM301 Implement MLOps practices with Amazon SageMaker

This session will showcase MLOps practices and how they can be implemented using AWS infrastructure. SageMaker has more and more features that are useful for both research and production so this session is helpful to keep track of them.

AIM202 Build, train, and deploy ML models using Amazon SageMaker

This session provides more of introductory information about how to get started with building an end-to-end ML pipeline using SageMaker. It’s useful if you want to learn what are the different services under the SageMaker umbrella and how they could be used.

MAD201 Containers and AWS Lambda: Choosing optimal compute for modern apps

One of the more expensive parts of ML/DL pipelines is computing. Depending on your needs you may need to use serverless (e.g. if you have to handle peak loads). Previously AWS Lambda had a 250MB package limit which limited applications where it could be used. With the 10GB, docker image limit which was introduced at Reinvent 2020 more ML applications can start to benefit from serverless computing.

CMP304 Selecting and optimizing Amazon EC2 instances

Similar to the previous session, this one will help to better understand the “menu” of different ways of using compute and how they can be optimized for your application.

DEV202 Continuous everything: Moving to DevOps on AWS

ML infrastructure usually needs to be flexible in case you want to change preprocessing, training, or prediction parts. One of the ways to do it is by organizing a CI/CD pipeline for your application. This session will help to learn how to do it using AWS infrastructure.



Rustem Feyzkhanov

I'm a staff machine learning engineer at Instrumental, where I work on analytical models for the manufacturing industry, and AWS Machine Learning Hero.