Title

AWS re:Invent 2022 - ML anywhere: Single pane of glass for cross-Region & on-premises AI/ML (PRT066)

Summary

Domino Data Lab provides a unified platform for data science work, enabling self-service access to compute resources and data, collaboration tools, and model deployment as an API without the need for IT intervention.
The platform also offers monitoring for deployed models to ensure they remain effective and valid.
Cross-region and hybrid deployments are essential due to data sovereignty and privacy laws, cost considerations, and the need for reliable service availability.
Domino's solution allows for seamless collaboration and governance across different regions while adhering to local data privacy laws.
The platform operates through a control plane (central management) and data planes (local processing) in different AWS regions and on-premises setups.
Domino integrates with various AWS services, runs natively on EKS, and can deploy models to SageMaker.
The platform is available for purchase on the AWS Marketplace and can be easily set up in new regions using EKS and a helm chart.

Domino Data Lab's platform addresses the challenges of international data privacy regulations by enabling localized data processing while maintaining centralized control.
The platform's ability to integrate with AWS services and run on both AWS and on-premises infrastructure provides flexibility and scalability for global organizations.
The use of EKS and helm charts for setting up new instances simplifies the expansion process for companies needing to deploy AI/ML workloads in multiple regions.
The ability to deploy models as APIs and monitor their performance reduces the operational burden on data scientists and IT teams, allowing for a focus on innovation and development.
The platform's design for collaboration and governance helps ensure that data access complies with local regulations, which is crucial for multinational companies.
Domino's approach to machine learning operations (MLOps) emphasizes cost efficiency by allowing data scientists to choose the appropriate compute resources for their tasks, potentially reducing unnecessary expenses.