11-11, 15:30–17:00 (America/New_York), Music Box (5th floor)
MLOps is a straightforward yet enveloping concept. Implementing MLOps standard practices can be challenging in small groups or companies. What are the bare minimum MLOps pieces you need to reap the benefits of MLOps and show ROI? We discuss a practical approach to building MLOps in small groups with a focused example.
MLOps is evolving into the common approach to ML lifecycle management. The core problem
addressed is how data science groups can move their models into production robustly.
Building out an MLOps system involves reproducible pipelines, monitoring, versioning models,
data versioning, and more. Do we need all of those things?
Typical MLOps best practices are often discussed in the context of large corporations, trillions of requests, and petabytes of data. Most organizations are not at this scale yet can benefit from aspects of MLOps. We discuss using a subset of MLOps with a real-world example addressed with Kedro.
No previous knowledge expected
Brian heads up the machine learning group at Shift5. He is a data scientist and entrepreneur with more than 25 years of experience across a variety of fields from NLP and speech processing to large-scale anomaly detection. Previously, he co-founded RuleSpace and served as chief architect, which was acquired by Symantec Corp. Brian received his Ph.D. in Computer Science from the Oregon Health & Science University of Portland, Oregon, a Masters in Manufacturing Engineering from Ohio University, Ohio, and a BA in Applied Mathematics and Computer Science from Hiram College, Ohio.