Highlights from Kubeflow Summit — October 2019
Last week I had a pleasure to be a speaker on the Kubeflow Summit organized on Google Campus in Sunnyvale, CA. It was a unique experience to meet community around 100 Kubeflow users and contributors, listen to exciting talks, and to share my and team’s expertise on moving ML pipelines to Google Cloud Platform.
The topics discussed that day covered:
- Continuous integration/continuous development (CI/CD) for Machine Learning workflows.
- Use cases from many companies (LinkedIn, Spotify, Bloomberg, OpenX, Cisco, GitHub, RedHat, Microsoft) and the scientific world (CERN).
- Building ML pipelines using TensorFlow Extended (TFX) and Kubeflow as a runtime environment.
- Security for Kubeflow cluster using Athos.
- On-premise Kubeflow installation.
- Scalable hyperparameter tuning using Katib.
An average number of weeks from idea to production
I was most impressed by the Spotify use case, where Josh Baer presented examples of how Kubeflow and TFX improved their ML workflow. Also, how they track this change in time using a dedicated metric: an average number of weeks from idea to production.
Distributed TensorFlow
The second fascinating talks were the use of cases from Bloomberg:
Supporting Distributed Tensorflow with Kubeflow Operators and Introducing KFServing: Serverless Inference Across ML Frameworks describing Istio and canary deployments process.
Kubeflow on the finance market
It was also exciting to hear about the technical challenges of using Kubeflow in the finance market on the talk: Fintech: Porting a Custom ML Model to Kubeflow. Mainly, how to use Kubeflow in a secure way required by financial companies.
Learn more
If you want to learn more, slides will be available on the Github page:
https://github.com/kubeflow/community/wiki/Kubeflow-Summit,-October-2019-Day-1-Agenda
Video recordings should be available on Kubeflow youtube channel soon: https://www.youtube.com/kubeflow
Thanks!
I want to say thank you to Kubeflow Summit organizers for the well-prepared event and the community for a full day sharing experience.
It was a unique opportunity to be there and see how Kubeflow is growing!
Kubeflow is open-source software, so feel free to engage by contributing: https://github.com/kubeflow or joining a mailing list: https://groups.google.com/forum/#!forum/kubeflow-discuss
Want to work at OpenX? Check open positions at https://www.openx.com/jobs/