First Ivanisevic Award recipient Samuel Drews researches fairness properties of automated decision-making

Computer Sciences PhD student Samuel Drews is the first recipient of a new award for dissertators: the Ivanisevic Award. This award was created by Dr Igor Ivanisevic (CS MS ’98; PhD 2000) and Dr Albena Ivanisevic (PhD, Chemistry, 2000). As UW-Madison alumni from two different departments, they decided to split the award between the two, resulting in a somewhat unusual structure. It is awarded to a dissertator in Computer Sciences for two years, and then one in Chemistry for two years, then Computer Sciences again, and so on.

Igor Ivanisevic describes their motivation for creating the award: “We wanted to pay forward UW for the happy times we had as doctoral students, and we know firsthand that getting support for students is one of the biggest challenges for any department.” The Ivanisevics hope that the award will allow recipients to “worry less about finances and more about education and research.” Albena is a professor, so they also know the challenge faced by faculty in finding financial support for grad students. “We are hoping this may help them out as well,” says Igor.

Drews is working on a PhD in programming languages. He says, “a substantial part [of programming languages research] involves the following: people write programs that do something—how can I write programs that automatically reason about what those other programs do?” A classic example, according to Drews, “would be trying to write a program that takes some other given program as input and automatically proves whether that given program could try to divide by zero (which would cause it to crash).”

Drews’s thesis involves “applying techniques from programming languages research to try to reason about novel concerns related to automated decision-making.” As an example, Drews says “a bank might use machine learning to obtain a model that tries to determine whether a loan applicant is likely to default on a loan, and they accordingly deploy this model (or ‘the algorithm,’ as the media tends to call it) to decide whether to give people loans.”

There are growing concerns about the fairness of these algorithms. Drews is working on creating a mathematical formalization that “allows one to express certain fairness properties, e.g., that the outcome of an applicant receiving a loan should be independent of the applicant’s race.” He continues, “Part of my research has involved (1) can we (automatically) prove whether or not a given program satisfies a given fairness property, and (2) if it doesn’t, can we (automatically) ‘repair’ the program by making some minimal change to it such that the repaired version does satisfy that property (while still being useful).”

According to Drews, the Ivanisevic Award will drastically improve the quality of his life over the next few years. “I think there is a lot of value in enabling graduate students to maintain a more even, healthier work-life balance,” he says, and Igor Ivanisevic agrees. “Reducing stress (financial or otherwise) and improving mental health were two of the main motivating factors behind the award,” Ivanisevic says.

Drews notes, “Staggering numbers of graduate students nationwide experience a lot of unique-to-grad-school stressors and grapple with difficult resultant mental health challenges.”  With their award, the Ivansevics are hoping “to help reduce [dissertators’] stress levels – [dissertators] have enough to worry about with academics and research” and take a step toward improving the mental health of Computer Sciences and Chemistry department PhD students.