Productive Machine Learning and Deep Learning Projects
summary
Platform design that decouples back end and front end and spawns language-specific containers (Python, R, Scala) with GPU access via an open wrapper for NVIDIA docker and orchestration on Kubernetes or Swarm automates training, testing and deployment. Projects use git-backed versioning for notebooks, data and models, trigger retraining from S3 webhooks, and let teams reuse microcontainers so models deploy fast rather than rebuilding stacks — inspired by Jupiter hub and packaged as 3Blades hub. Challenges include massive legacy fragmentation (Halliburton >200 ERPs).
AI-written summary · about
chapters (13)
- 0:00Intro
- 1:21Background
- 4:31Challenges
- 5:32The old way
- 6:53The Venn diagram
- 7:38The data gurus
- 9:07The new deliverable
- 10:44Nvidia Docker
- 12:26DevOps
- 14:31Solution Architecture
- 15:40Three Blades Hub
- 18:35Conclusion
- 20:15Questions
related talks
42mThe Challenges Managing a Kubernetes-Based Machine Learning Infrastructure
21mMastering the DevOps Machine Learning Pipeline
39mOperating Deep Learning Pipelines Anywhere Using Kubeflow
42mDeep learning beyond the learning
28mProductionizing Machine Learning Models at Scale with Kubernetes
46mDeploying scalable Machine Learning models with Kubernetes
17mDevOps for Private AI: A Scalable Platform Using Kubeflow
22mBuild and Monitor Machine Learning Services in Kubernetes