Defending a Ph.D. is stressful. Starting a new faculty position is also stressful. Starting a new faculty position in a foreign country four days after defending? Ask Filippo Ascolani about it.
The new assistant professor of Statistical Sciences — who comes to Duke fresh out of his Ph.D. at the Università Bocconi, in Milan, Italy — doesn’t feel totally out of water, though.
“I had the opportunity to meet a lot of people from the department in conferences around the world,” he said. “I met some of the Ph.D. students and professors here many times over the last four years, so in a way it was like coming somewhere known.”
Ascolani’s award-winning research focuses on a class of statistical models known as nonparametric Bayesian statistical models. These models deal with data that doesn’t follow a pre-determined distribution, such as a bell curve. They grant a high flexibility, allowing researchers to avoid stringent assumptions when describing the phenomena of interest.
Bayesian statistical models are widely used in fields as disparate as economics, biology and biomedical sciences. They aren’t without constraints, though. Like an architect problem-solving a blueprint, Ascolani develops theoretical models to address these constraints. His focus is a model’s ability to deal with heterogeneous data, i.e., data that hasn’t all been collected in exactly the same conditions.
As an example, Ascolani asks us to consider a clinical trial being run in two separate hospitals. While many of the trial’s parameters — drugs used, in which frequency, in what type of patients, etc. — are kept constant in both places, others — doctors and nurses in charge, room conditions — invariably differ. Typically, data from each hospital would be kept in its own separate bucket to account for this heterogeneity. To combine them, current methods often require different subgroups of data, or buckets, to correlate positively with the others.
Data doesn’t always correlate positively, though. Take stocks and bonds: they may typically fluctuate together in the same direction (i.e, have a positive correlation), but sometimes they go their separate ways. Being able to use heterogeneous data to predict the market’s behavior while taking into account both periods of positive and negative correlation isn’t a trivial statistical problem.
To complicate matters further, buckets of data are seldom complete: Maybe one hospital lost a week of data due to a power loss, or the other had no patients in a certain age group. Being able to borrow information from one group into the other would allow these gaps to be plugged, making the analysis more robust. The models developed by Ascolani and his coauthors do just that.
Ascolani — who has a background in mathematics through bachelor’s and master’s degrees at the Università di Torino — specifies that his research is currently focused on theory and computation. “I want to understand how a certain model (and the associated algorithms enabling statistical inference) behave if you have a lot of data,” he said, “or if it’s going to be able to explain a certain feature of the data, provided it's had enough information.”
At Duke, he is excited by the possibility of working in close collaboration with researchers wrangling messy real-world datasets.
“If you work with applied scientists you get directly at the problems people want to solve, and which features they want from a statistical model,” Ascolani said. “The biomedical sciences are typically interested on this kind of borrowing of information. There are also many difficult and interesting problems when dealing with economics, econometrics, macro econometrics, etc.”
The potential for exciting collaborations isn’t the only thing Ascolani liked about Duke.
As he gets used to life in the United States and in Durham, Ascolani finds challenges — he hadn’t had to drive since 2019 — but also joys, such as the greenery and the abundance of running routes.
“I was a little worried when moving from Europe to the U.S., since it is a completely different environment,” he said. “I have to say that the fact that Durham is such a nice, enjoyable city made it much easier for me to settle in and slowly learn about the features of U.S. life.”
The best surprise was right here on campus. “I read a lot, and I still enjoy reading in Italian, but books in Italian are very difficult to get,” he said. “But the Duke library has a large corridor of books in Italian. That was very unexpected, and also very appreciated.”
Statistical Science
