Machine Learning Platforms at Uber with Mike Del Balso - #115

Infrastructure

Implementing machine learning infrastructure at scale presents numerous challenges for organizations. Mike Del Balso highlights the importance of integrating robust infrastructure to support machine learning systems, emphasizing that a great model alone is insufficient without the necessary logging and monitoring systems 1. He notes that the complexity of machine learning platforms requires a significant engineering investment to ensure stability and scalability, particularly in high-demand environments like Uber's ad systems 2.

Even if you have a great model, there's a lot more infrastructure around your machine learning system that you need to integrate with.

--- Mike Del Balso

This integration is crucial for maintaining model accuracy and performance over time, preventing models from becoming stale or ineffective.

   

Michelangelo

The Michelangelo platform at Uber plays a pivotal role in deploying and managing machine learning systems efficiently. Mike Del Balso explains that Michelangelo was developed to address the need for a unified platform that supports the entire data science workflow, from exploration to production 3. This platform allows teams to share features and avoid duplicating data pipelines, enhancing collaboration and efficiency 3.

Our goal is to support everything from the exploration side of the data science workflow all the way to production.

--- Mike Del Balso

By providing a comprehensive infrastructure, Michelangelo helps teams at Uber, such as those predicting ETAs or managing Uber Eats, to deploy and maintain machine learning models at scale 2.