Cloud Native Machine Learning Systems at Day Two and Beyond
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Thursday, 2020-11-19· at 14:55· Intrado Virtual Event Platform,·
summary
Kubernetes becomes the default infra for ML but it leaves gaps around reproducibility, interactive velocity, and model monitoring; expect to use CI/CD (Tekton) and image builders like source-to-image to produce certified images, handle builder pods with GPUs or offload huge trainings, and instrument pipelines with drift detection using tools like alibi explain and Great Expectations. Concrete needs: versioned data and feature encoders, rollback-able library stacks, automated model-image generation, and multi-point monitoring to catch silent prediction drift.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:01Agenda
- 2:16Machine Learning Systems
- 4:54Hidden Technical Debt
- 5:36Reproducibility
- 5:52Feature Engineering
- 7:32Library Stack
- 8:18Explainability
- 9:13GDPR
- 9:43High Velocity Data Science
- 12:42Delivery Challenges
- 14:01Inputs Outputs
- 20:48The Big Picture
related talks
06mEvolving with Kubernetes: Embracing Model Ops
12mKeynote: Building Enterprise AI/ML Platform on Kubernetes
21mMastering the DevOps Machine Learning Pipeline
26mMachine Learning Meets K8s
44mDevOps for Machine Learning - why is it different?
42mKubernetes for Machine Learning: Managing and Scaling AI Workloads
39mContinuous Delivery for Machine Learning — The Future of MLOps
42mThe Challenges Managing a Kubernetes-Based Machine Learning Infrastructure