/ GitOps for Reproducible Machine Learning

Description

Git has revolutionized how we develop and manage software projects, enabling efficient collaborative development cycles. Adopting Git to manage machine learning projects was intuitive but introduced challenges never encountered before. Support for scalable datasets, experiment management, and reproducibility revealed blind spots that made its usage in ML projects an uphill battle. In this session, I'll share a fully reproducible GitOps methodology for ML, developed over the course of three years of research. You will learn how to utilize Git's versioning capabilities and non-linear workflow to unlock full reproducibility and iterability, parallelization of work on all project components, as well as reduce the iteration time from research to production. We'll extend Git's functionality to cover the needs of ML projects while being tools agnostic and focusing on the core capabilities required.

Session 🗣 Intermediate ⭐⭐ Track: AI, ML, Bigdata, Python