"We Have No Idea How Models Will Behave in Production Until Production": ML Ops
a day ago
- #MLOps
- #Ethnographic Study
- #Machine Learning Engineering
- Organizations rely on machine learning engineers (MLEs) to deploy and maintain ML pipelines in production.
- MLOps requires proficiency in both data science and engineering due to models' reliance on fresh data.
- Ethnographic interviews with 18 MLEs reveal a workflow involving data preparation, experimentation, evaluation, and continual monitoring.
- MLEs collaborate with data scientists, product stakeholders, and peers using tools like Slack and ticketing systems.
- The 3Vs of MLOps are introduced: velocity, visibility, and versioning, which are key to successful ML deployments.
- The study discusses design implications and opportunities for future work in MLOps.