NORDICS20 - Assets

Deep Learning on AWS

Amazon Web Services Resources EMEA

Issue link:

Contents of this Issue


Page 31 of 50

Amazon Web Services Deep Learning on AWS Page 27 There are also AWS pipeline components for Kubeflow that integrate with Amazon SageMaker and other AWS services used for data cleaning and transformation, such as Amazon EMR and Amazon Athena. This approach is for customers who want a unified control plane (unifying their microservices architecture with their AI/ML service releases) but also want to leverage different AWS services, such as Amazon EKS, Amazon FSx for Lustre, and Amazon SageMaker that are best fit for job and can help with the undifferentiated heavy lifting. Patterns for Deep Learning at Scale In this section, we describe different solutions and patterns that you can use to scale out deep learning adoption in your organization. The options include fully managed, DIY using AWS services, or a hybrid approach. You can also extend these options using other AWS services to build a custom environment. Options for Deep Learning on AWS In this section, we discuss both fully managed and DIY platform as a solution approaches for deep learning on AWS. AWS offers multiple options that cover the full spectrum of solutions for deep learning depending on different levels of shared responsibility between AWS and the customer, as shown in the following figure. Figure 6: Deep learning solutions on Shared Responsibility Model spectrum The left side of the spectrum is Amazon SageMaker, fully managed service for deep learning, event-based, and provides a single-click build, train, deploy experience. On the right side of spectrum is a fully customized deep learning solution that is self- managed, manual and involves multiple steps to build, train, and deploy. In between these options is a partially managed solution where you can simplify infrastructure management by using managed services, such as Amazon EKS and you can deploy self-managed Kubeflow for deep learning workflow.

Articles in this issue

Links on this page

view archives of NORDICS20 - Assets - Deep Learning on AWS