‎

A grad’s view: Solving real problems essay professional

This is a re-port of an article I wrote for the MassMutal blog: https://blog.massmutual.com/post/a-grads-view-solving-real-problems.

Just this past May, I graduated from the University of Vermont and the Computational Story Lab research group to work as a senior data scientist with MassMutual. While there are many differences between being a lifelong student (read: PhD student) versus working in industry, the overlap is huge and I have been able to settle right in at MassMutual.

Differences

The biggest difference is what defines value. Research in machine learning is valued by novel algorithms and marginal improvements in classification accuracies. But, in business, our data science efforts and machine learning algorithms provide value by making improvements to the customer experience.

While spending months of effort developing an improved algorithm and moving accuracy from 90 percent to 92 percent would be heralded in academic journals, moving a process from 50 percent to 90 percent is where all the gains are. (Related: Data science and MassMutual).

Overlap

One of the biggest reasons that things are so similar is that the Data Science team at MassMutual has co-opted and applied many of the practices that work well at universities. I found it easy to feel at home with weekly lab meetings, a culture of learning from peers complete with a mentorship structure, and an interval review process for project write-ups akin to journal papers (without any, ahem, rude reviewers).

Add this to the fact that the Computational Story Lab’s use of machine learning and computation are used as tools to study interesting problems, and it’s a match.

What universities could learn

On the flip side, what I have learned in my few months at MassMutual could go a long way to improving research.

We have a Data Engineering team to support the computing and data platform (excitement from Day 1), track our work in a way that helps keep projects moving, follow a consistent coding style across the team, and ensure that our models are reproducible. The latter part, building reproducible models with a consistent project structure, has been an effort that I and another team member, Paul Shearer, worked on and implemented recently.

Finally, while our focus is on solving business problems, that isn’t to say that there isn’t any science going on. Our models to predict longevity utilize new methods that I still haven’t wrapped my head around. We tweak state-of-the-practice methodology for tricky problems, and use experimentation to track the results of our models and improve them.

We’re bringing applied machine learning to areas across the company, and it’s an exciting place to be!