Kubeflow Intro – Michał Jastrzębski & Ala Raddaoui, Intel
summary
Kubeflow packages running training and serving on Kubernetes using real components: TFJob/TensorFlow Operator to spawn distributed training, Jupyter notebooks for development, Pachyderm for data pipelines, plus Rook PVs and GCS for storage. Demo deploys to a 17‑node cluster, builds a Docker image, and launches a convolutional model training on MRI brain scans (training takes hours). Model serving is automated but the demo serving path failed, highlighting early‑stage instability and need for contributors.
AI-written summary · about
chapters (9)
- 0:00Introduction to Kubeflow
- 1:21Bridging Data Science & DevOps
- 3:47The Model Lifecycle
- 5:03Infrastructure & Architecture
- 5:47Installing Kubeflow
- 8:17Model Development & Data
- 11:58Training the Model
- 16:31Serving Models in Production
- 18:51Q&A and Contribution
related talks
21mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
29mKubeflow: TensorFlow on Kubernetes
14mKubeflow: Where ML Meets Kubernetes
35mKubeflow: Data Scientist ready?
46mDeploying scalable Machine Learning models with Kubernetes
46mKubeflow, TensorFlow on Kubernetes
31mKubeflow - Data Science on Steroids
30mKubeflow, Your ML Tools on Kubernetes