Story|

Making Way for the Future of Machine Learning

A new course at the University of Chicago prepares the next generation of data scientists to use artificial intelligence in the enterprise realm.

Machine Learning Actionable Frameworks

Eighty-five percent of artificial intelligence solutions will fail.

This sobering statistic, delivered by research and advisory giant Gartner, Inc., is in large part a matter of implementation: although data scientists may produce good individual work, that work, more often than not, cannot be reconciled within the larger structures in place in their organizations.

To combat this rampant failure, the Master of Science in Analytics (MScA) program at the University of Chicago has launched a new course in Machine Learning Operations (MLOPs). The field of MLOps is nearly as new as the course. Its structure is cribbed from that of DevOps, the set of practices that elevate and regulate software development.

“Software engineering is a mature field, while data science is relatively new and needs its own DevOps for all machine learning and AI applications,” says Dr. Greg Green, Executive Director of Analytics at UChicago. “The idea was to leverage best practices from more mature disciplines and create a parallel field so that the work of data scientists fits within the ecosystem of a corporation and is scalable, not just applicable to a single-point solution.”

“Most master's-level data science programs are teaching students how to solve problems, but not how to implement their solutions within an ecosystem of other solutions. The MLOps course is designed to show students the platforms and tools that enable them to implement their solutions in a collaborative and scalable way.”

Dr. Greg Green, Associate Senior Instructional Professor, Director MScA Program

“The reason I came up with the course is that there is a gap between what we teach in the classroom and what’s expected in the industry,” says Dr. Arnab Bose, MLOps instructor and Chief Scientific Officer at Abzooba, a data analytics company. “We don’t teach the complete workflow, but the solutions that students learn and the models they build can actually be deployed in production systems. This course fills the gap between what they learn and how they can implement it in an industry setting.”

This mission sets the Master of Analytics program at the University of Chicago apart from other data science programs, says Green. “Most masters-level data science programs are teaching students how to solve problems, but not how to implement their solutions within an ecosystem of other solutions,” he says. “The MLOps course is designed to show students the platforms and tools that enable them to implement their solutions in a collaborative and scalable way. When a graduate of another master's program lands a job, the first thing they’re doing is learning new tools. We want them to hit the ground running and accelerate their time to full value and maximize their early-stage contributions.”

Students in classroom working on laptop

Ready to lead the field of machine learning?

Start your application today and join our cohort of data science professionals. 

Apply Now

A large part of bringing students up to speed involves exposing them to the machine-learning platforms most commonly used in enterprise production. “In this industry niche, there are some heavy-hitters that have been around for a few years: DataRobot and Dataiku,” Bose says. The course also teaches newer niche platforms like Allegro AI, Lityx and xpresso—the latter produced by the company where Bose serves as Chief Scientific Officer. So impressed was he with the course’s ability to address corporate needs that the CEO of Allegro Trains let Bose use his platform in addition to guest speaking on a complimentary basis in the initial course launch.

“The class addresses such a fundamental need in the industry that it serves those who want to be hands-on, individual contributors, those who want to manage teams of data scientists, and those who plan on being consultants,” Bose says. “Real-life data is continuously changing, so you need to know how to use the right models.”

Additional Stories