Keynote: Kubeflow Endoscopy
summary
Medical AI team applies Kubeflow and Katib to gastric endoscopy classification and reports concrete wins: hyperparameter tuning drops from 5 months to 2 weeks (about a 10x speedup), model serving shrinks from 1.5 months to 2 weeks, and the workflow becomes reproducible and triggerable. Implementation uses Jupyter notebooks for development, Katib without code changes to run extensive searches on TensorFlow/Keras models, Pipelines to orchestrate data-prep → tuning → serving, and is moving toward KFServing for production.
AI-written summary · about
related talks
15mKeynote: Machine Learning on Kubernetes Made Easy With Kubeflow
27mKubeflow for enabling AI Powered Drug Discovery and development...
10mKeynote: Machine Learning using Kubeflow and Kubernetes
21mKeynote: A Tale of Using Kubeflow to Make the Electricity Smarter in China
14mLightning Talk: Leveraging Kubeflow for Next-Generation AI Healthcare Solutions
35mBuilding a Medical AI with Kubernetes and Kubeflow
49mTutorial: From Notebook to Kubeflow Pipelines to KFServing: the Data Science Odyssey
36mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)