By Elea Levin
Of the paper entitled “Proving Data-Poisoning Robustness in Decision Trees,” D’Antoni says, “This work is the first of his kind, as it’s trying to provide formal proofs of whether a machine learning model is robust with respect to potential errors in the training data set.”
D’Antoni credits Samuel Drews as “the pioneer behind this idea, which he proposed as part of a course project here at UW-Madison. This was also the last piece of his dissertation, and it’s great to see that it’s getting such an important recognition.”
“It’s a surprising idea with great implications,” says Albarghouthi. “While the original focus of the work, as conceived by Dr. Drews, was on trustworthiness of machine learning, over the past year we have been applying it to understand how historical and other forms of bias in data sets affect the decisions made by machine learning models.”
SIGPLAN is the ACM Special Interest Group that focuses on exploring concepts in programming languages composed of programming language developers, educators, implementers, researchers, theoreticians, and users.
Papers that receive this honor are selected by a committee that represents SIGPLAN’s major conferences and elected officials. The committee selects honorees from papers that are presented at their conferences, and selected papers are “of high quality and broad appeal.”
The paper will also be submitted to the CACM Research Highlights editorial board for their consideration.
D’Antoni earned Bachelor’s and Master’s degrees in Computer Science from the University of Turin. He went on to earn a PhD in Computer Science from the University of Pennsylvania. He is the recipient of a Microsoft Research Faculty Fellowship, an NSF Career Award, and a Facebook TAV Research Award.
Albarghouthi received his PhD in 2015 from the University of Toronto and has earned an NSF Career Award and Facebook Probability & Programming Research Awards.
Drews received his PhD from UW-Madison in 2020 and now works as a research scientist at Facebook.