Teaching Causal Relations to Machines: Bocconi Summer School in Advanced Statistics

THIRTY PHD STUDENTS HAVE SPENT TWO WEEKS OF LESSONS AND EXERCISES AT LAKE COMO. WITH A PREDICTION COMPETITION WON BY A BOCCONIAN

Tommaso Rigon, a second-year student of Bocconi’s PhD in Statistics, won the Prediction Competition at the Bocconi Summer School in Advanced Statistics and Probability. The course, which ended on Saturday after two weeks of lessons and exercises, was held at Villa Grumello, on Lake Como.
 
The thirty students participating in the competition were given a first dataset of credit card transactions, some of which were flagged as fraudulent, and a second dataset that only specified the features of another credit card transaction series. Their task: to predict, in a few hours, which transactions in the second series were fraudulent. "I tried to apply different models to the first set of data, and I picked one that I thought was stable, simple and functional," says Rigon. "With more time available, the strategy would have been different, I would have looked for greater accuracy, but in these predicaments I felt it was the best thing to do."
 
"The summer school had thirty participants, selected from PhD students across Europe and, in one case, from the United States," says Sonia Petrone, co-director of the initiative. The topic of the summer school was Statistical Causal Learning, which is one of the cutting-edge aspects of machine learning. "Machine learning allows machines to get more accurate predictions," says Petrone, "but statistical causal learning wants to teach machines how to identify the cause-effect relationships of the phenomena to predict."
 
Compared to other summer schools in Statistics, Bocconi’s stands out for two aspects: the audience, consisting mainly of PhD students instead of PhD graduates or researchers, and a structure that preserves the rigor of doctoral education. The summer school lasts two weeks, and not a few days, and the activities are very structured, with lessons, exercises, group work, tutorials, coding and, of course, a competition. "In short," Petrone concludes, "it aims to be the Bocconi contribution to a European doctoral network."


Tommaso Rigon

by Fabio Todesco