MLOps is an emerging field of best practices for businesses to run ML workflows in production. There are multiple challenges associated with MLOps which makes it different from traditional DevOps. From monitoring where you may need to monitor model performance and change in data to cost and scale optimizations where you need to optimize for GPU idle time. New tools and approaches which help to tackle the challenges appear every month so here are the top 5 AWS re:Invent 2020 talks to help you stay in the loop of MLOps in the AWS cloud.
re:Invent is free and virtual this year so now it’s a lot easier to access the content. One of the best ways to find and save all the sessions is to use Cloud Pegboard.
From POC to production: Strategies for achieving machine learning at scale
The session for executives and managers to learn how to organize ML at scale in their companies and how to solve challenges like MLOps, data governance, and knowledge sharing.
Implementing MLOps practices with Amazon SageMaker
The session for data scientists and IT professionals which shows how to solve common MLOps challenges using SageMaker and how to implement each part of your end-to-end ML pipeline in the cloud.
Detect machine learning (ML) model drift in production
One of the challenges with ML models compared to code is that they can become outdated because it makes predictions on the new data which is very different from its train dataset. This session showcases SageMaker tool which provides a way to detect drifts in predictions.
Secure and compliant machine learning for regulated industries
ML pipeline heavily operates with data during both training and predictions. If your data is very sensitive then organizing the pipeline may become a challenge. This session shows patterns and architectures with Amazon SageMaker which tackles the challenge.
Scaling MLOps on Kubernetes with Amazon SageMaker Operators
One of the challenges with scaling ML part of the infrastructure is that you may need to scale both training and inference pipelines across multiple regions and availability. This session shows how to do it by using Kubernetes with Amazon SageMaker Operators.