Drunken Idea Leads to Wild ML Adventure & ML-Ops: The Quest for Full-Blown SCM
Drunken Idea Leads to Wild ML Adventure By Liza Katz (Software Engineer)
The story begins with a night of excessive drinking in great company, as many great stories do. An interesting and provocative idea grabbed my attention. Weeks have passed, and I couldn't stop thinking about it, so I decided to give it a shot.
This talk tells the story of my journey pursuing that idea into the ML realm: full of twists and turns, late night debugging sessions, kind and helpful people, technical challenges, frustrations, and many many photos I'll never be able to forget.
ML-Ops: The Quest for Full-Blown Source Control By Itai Admi (Team Leader @ Treeverse)
As the industry slowly progresses towards full-blown ML-Ops platforms - implement design, model development and operation of machine learning. Automating data science operation with CI/CD has many requirements that as an industry we almost solved. For example Feature Store with Tekton, ML Metadata Store with Neptune, and ML Flow orchestrator with Kubeflow. Having said that, we are still missing the piece of full blown source control. Which includes, version control for Code, Data, and ML Model artifacts. Using Github we can version control our code and ML artifacts pretty easily. When trying to version our data, we hit the wall of scalability and security. Our next challenge in tracking the ML-Ops journey is understanding how to fully version control our data and connect it with the rest of the machine learning automation efforts. In this session, you will learn how to track your MLOps journey, the tools that proved themselves in the industry and take full advantage of OSS tools for versioning your data.