Caroline Uhler and I are co-directors of the new Eric and Wendy Schmidt Center within the Broad Institute of MIT and Harvard, which seeks to catalyze research at the interface of the life sciences and the data sciences. While Caroline is someone I now talk to every day, we had actually never met until June of last year.

Here's how we started working together: As the idea of the Schmidt Center evolved, the question emerged as to who should be its leaders. Eric Lander (who we talked to last season on Theory and Practice) and Genentech's Aviv Regev (a guest coming up in this series), knew both Caroline and me independently, but Caroline and I did not know each other. Eric and Aviv decided to play a bit of a matchmaking game, given our very different but complementary strengths. They said, 'Caroline is a rising star of machine learning, and a dedicated educator and mentor. Anthony is a doctor, and more operationally-focused'. In some ways, I think it's fair to say they saw us as a bit of a hammer and nail pairing, a theme of our show!

I admire Caroline for many reasons. First, she is the rare and wonderful combination of being both brilliant and humble. Many academics of her stature live their life on a quest for intellectual immortality; Caroline loves what she does, and wakes up every day excited that she gets to do it. Moreover, Caroline is someone who's not just trying to beat the latest benchmark on a dataset, but is really thinking hard about the mathematical underpinnings of why machine learning works. Finally, she's a dedicated educator and mentor, and her students love working with her. As I mentioned in this episode, I'd love to go back and do a second Ph.D. with her!

Caroline Uhler Caroline Uhler, MIT

Caroline's research in machine learning has largely focused on three areas: The first is causal inference (what does one have to believe about observational data to say that one variable is a cause of another?). The second is representation learning (taking datasets and mapping them into new spaces, where their structure becomes clearer.) These techniques are both being used to explore how to repurpose existing drugs to treat Covid-19.

The third area is a bit more mysterious, and is a field that Caroline and a few of her collaborators really brought to the forefront. It's called multivariate total positive distributions of order 2 (MTP2). In life, you often have sets of variables with positive correlations. Caroline and her colleagues showed that these often have a remarkably simple underlying structure. This turns out to be a powerful approach to understanding complex systems.

Caroline is not just a machine learning researcher, however. She has also turned herself into a deep biologist focused on gene regulation. Even now she is building an experimental lab at the Broad.

Getting to know Caroline these last few months has been one of the great joys of my professional life. I think you'll see why during this episode of Theory and Practice.