DataFramed #67 Operationalizing Machine Learning with MLOps
Jul 26, 2021
Alessya Visnjic, CEO and co-founder of WhyLabs, brings her expertise from Amazon and the Allen Institute for AI to discuss MLOps. She tackles the unique challenges data teams face in operationalizing machine learning and explains the differences between MLOps, DataOps, ModelOps, and AIOps. Alessya emphasizes the importance of starting early in your MLOps journey and shares insights on fostering a strong data culture. She also introduces Ylogs, an open-source library aimed at standardizing data logging and enhancing AI reliability.
AI Snips
Chapters
Transcript
Episode notes
Amazon Experience
- Alessya Visnjic joined Amazon in 2008 when it was still an emerging company.
- She witnessed the explosion of DevOps culture and worked on large-scale machine learning deployments, like demand forecasting for Amazon retail.
Continuous MLOps
- MLOps aims for continuous and consistent model performance, mirroring DevOps in traditional software.
- Machine learning models must adapt to new data continuously, unlike static software rules.
MLOps vs. AIOps/DataOps
- AIOps uses AI for IT issues, while MLOps operationalizes AI/ML applications.
- DataOps focuses on data pipelines, while MLOps focuses on ML pipelines, with overlap.