Dive into a comprehensive 55-minute conference talk from MLOps Community Meetup #119 featuring Aurora Innovation's senior engineers discussing how they accelerated machine learning model development for autonomous vehicles using Kubeflow. Learn about the evolution of Kubeflow infrastructure, pipeline building, developer experience, and the benefits of using pipelines. Explore the process of adopting Kubeflow organization-wide, including local deployment, multi-user setup, authentication, high availability, and lessons learned. Gain insights into Aurora's AV development workflow, pain points, and solutions implemented through Kubeflow components. Discover how the team leverages various tools like Batch API, Sagemaker, and GitHub PR comments in their pipeline design. Get recommendations for those starting with Kubeflow and understand its comparison with other competitors in the MLOps space.
Accelerating ML with Kubeflow for Autonomous Vehicles
Overview
Syllabus
[] Introduction to Ankit Aggarwal, Vinay Anantharaman, and Maurizio Vitale
[] Team Aurora's Introduction
[] Agenda
[] Introduction
[] Aurora Innovation is hiring!
[] AV Development Workflow
[] Pain points
[] Designing ML Orchestration Layer
[] Kubeflow Overview
[] Kubeflow Pipelines
[] Kubeflow Components Overview
[] Batch API
[] Sagemaker
[] GitHub PR Comments
[] Slack Notifier
[] Pipeline design
[] Developer Workflow
[] End-to-end Pipeline
[] Unified UI
[] Benefits
[] Kubeflow Infrastructure
[] Local Deployment
[] Multi-User
[] RDS
[] IAM
[] Vault
[] S3
[] User Autehntication
[] Groups
[] High Availability
[] Lessons Learned
[] Pipeline Runs
[] AVG Runs Per User
[] Pull Requests Verified
[] Back-end storage Opt-out
[] Other main competitors
[] Happiness on Kubeflow
[] TFX Tensorflow Serving
[] Track database access back to an individual user differentiating between individual users
[] Open-source Kubeflow
[] Open-source Community aligning values
[] Sagemaker components of Kubeflow
[] Kubeflow pipeline syntax
[] Recommendations to Starters
[] Vertex as a managed service
[] Deploying service to teams across the organization
[] Wrap up
Taught by
MLOps.community
