Building a Reproducible Machine Learning Pipeline With Kubernetes, Tensorflow, and Kuberflow

Ieva Vaišnoraitė-Navikienė; CTO at Attention Insight
SwissAI Machine Learning Meetup on "Deep Learning with TensorFlow"

MAIN TOPICS:
1. What are Google Kubernetes and Kubeflow and when should we use them?
2. How to use Kubeflow with Tensorflow to create reusable machine learning pipelines?
3. How to set-up and maintain machine learning pipeline versioning control?

ABSTRACT: Machine learning is an extremely repetitive process. In case to produce good predictions many steps and even sequences of steps are repeated over and over again. This is particularly important when we are talking about biodata which could be preprocessed in many ways. That adds even more complexity to the data analysis pipeline. Pipeline complexity issue could be tackled in many ways and the Google team suggests many tools that are targeting different aspects of it. In the presentation, I will show Kuberflow - one of the tools to structure containers into pipelines and in that way make the solution more portable and reproducible.

ORGANIZERS

SwissAI Machine Learning Meetup is one of the larges AI meetups in Switzerland, with regular meetings, great speakers invited from academia and industry and over 1600 members.

Pawel ROSIKIEWICZ, Founder of SwissAI, Main Organiser

Prof. Martin JAGGI, Head of Machine Learning and Optimisation Laboratory, EPFL

Veronica ALONSO KANONNIKOFF, Marketing & Communication Advisor

Chris PEDDER, Ph.D, NLP ML engineer & Copywriter

Luca PESCATORE Ph.D; Data Scientist & Particle Physicis

Onur YÜRÜTEN, Volunteer, Data Scientist

Salar Rahimi, Volunteer, MSc Student, EPFL

Keshav SINGH, Volunteer, MSc Student, EPFL

Clement CHAROLLAIS, EPFL, Camera Operator and Movie Editing

SPONSORS

École Polytechnique Fédérale de Lausanne (EPFL)

Innovaoud

Noe Lutz, Engineering Lead at Google AI - Private sponsor