Unified MLOps: Feature Stores and Model Deployment

Explore a revolutionary approach to MLOps in this 29-minute talk from Databricks. Learn how to scale machine learning models without increasing latency by combining a database, feature store, and machine learning. Discover Splice Machine, a hybrid database built on HBase and Spark, which offers a unique single-engine feature store and deploys ML models as database tables. Understand how HBase enables millisecond feature serving and prediction generation, while Spark facilitates complex training set creation and large-scale ML predictions. Gain insights from the speaker's experience at NASA's AI lab and various software companies. Watch a demonstration of Splice Machine's capabilities and its integration with Databricks through a simple JDBC connection. Delve into topics such as unified MLOps, feature store functionality, training set generation, and feature serving for efficient model deployment and scaling.

Intro
Challenges
Unified MLOps
Feature Store Demonstration
Databricks Demo
Training Sets
Feature Serving