Title

AWS re:Invent 2022 - [NEW] Improve ML governance w/deep control & transparency in SageMaker (AIM346)

Summary

• Ankur Mehrotra, GM for Amazon SageMaker, and Sai, a principal product manager, along with Dominic from Fidelity Investments, discuss ML governance and new SageMaker capabilities.
• Machine learning (ML) has moved from POCs and pilots to critical business decision-making at scale.
• ML governance is essential for ensuring privacy, quality, and compliance throughout the ML lifecycle.
• Complexity in ML processes is increasing due to more stakeholders, an exponential growth in models, and proposed government regulations.
• ML governance capabilities needed in ML systems include user onboarding, capturing decisions made by data scientists, and continuous monitoring of models in production.
• Amazon SageMaker has launched new features to address ML governance challenges: SageMaker Role Manager, SageMaker Model Cards, and Model Dashboard.
• Sai demonstrates how to use these new features, emphasizing ease of use and integration with existing AWS services.
• Dominic from Fidelity Investments shares their experience with SageMaker and the importance of ML governance for their operations.

Insights

• The transition of ML from experimental stages to production at scale signifies a maturation of the technology and its integration into core business functions.
• The involvement of multiple stakeholders in the ML process, including legal and compliance teams, highlights the importance of governance and the need for tools that cater to diverse roles.
• The exponential growth of models within enterprises underscores the challenge of maintaining oversight and governance at scale.
• Government regulations are driving the need for transparency in ML processes, necessitating features that can provide detailed documentation and audit trails.
• SageMaker's new features aim to simplify the onboarding of users, standardize model documentation, and provide a unified view of model performance and lineage, which are critical for governance.
• The demonstration of the new features by Sai shows a commitment to user experience and the practical application of governance in ML workflows.
• Fidelity Investments' use case illustrates the real-world impact of ML governance on business operations and the importance of tools that can support a large number of models and data scientists with a limited number of ML engineers.
• The presentation suggests that AWS is actively listening to customer feedback and evolving its services to meet the growing demands for ML governance in various industries.