How Feature Stores Can Accelerate Your Path to Scalable MLOps
Webinar Speaker: James Kobielus, Senior Research Director, Data Management
Date: Wednesday, January 19, 2022
Time: 9:00 a.m. PT, 12:00 p.m. ET
Features are individual measurable independent variables of a phenomenon to be predicted. When building a statistical model of that phenomenon, data scientists must select the features best suited to reducing the model’s computational cost and boosting its predictive accuracy.
Feature selection is a core responsibility for data scientists. It involves identifying which features in the data best predict some quantifiable outcome of interest. To the extent that data scientists can reuse previously discovered features in future machine learning (ML) projects, they can streamline and accelerate the process under which statistical models are built, trained, and deployed into production.
One key piece of data infrastructure in this regard is the feature store. This refers to a specialized data system and abstraction layer for storing, managing, accessing, serving, and reusing canonical feature representations within repeatable MLOps data pipelines.
Join TDWI’s senior research director James Kobielus on this webinar to explore the pivotal role of feature stores in improving the efficiency, effectiveness, and speed of today’s MLOps pipelines. He will discuss how these stores, as a “single version of the truth” for the features most predictive of ML model performance, can help data science professionals to:
- Automate feature discovery, computation, and transformation
- Share and explore alternative feature representations
- Serve features consistently across training and inferencing scenarios
- Deploy and iterate new features rapidly into production environments
- Enable rapid remediation of in-production model performance issues
- Support feature registration, tracking, versioning, and governance
Kobielus will be joined by an expert from Snowflake to discuss how today’s data scientists can benefit from the use of feature stores as part of their cloud-based MLOps pipelines.
Julian is a product marketer with experience in both high-growth startups and established enterprise software and infrastructure providers. Prior to Snowflake, Julian was part of Confluent during the fast-growth journey to IPO. He began his career at IBM where as a consultant, he advised large enterprises on their machine learning and analytics strategy. Julian holds an aerospace engineering undergraduate degree from Georgia Tech and an MBA from The University of Chicago. During his free time, Julian enjoys road biking around the San Francisco bay area.