When you apply for a credit card or a mortgage, who or what is behind the decision-making process that approves or denies your application? Increasingly, banks are using software algorithms to make decisions like these. But how fair are the algorithms? Will you be approved because of your stellar credit score, or declined because of the zip code you live in?
These issues of equitable algorithms and systems are being tackled by College of Information and Computer Sciences (CICS) faculty in an initiative chartered by Professor Gerome Miklau, called EQUATE (Equity, Accountability, Trust, and Explainability). “We formed EQUATE to bring together the broad set of CICS faculty already doing research in equity-related computer science including algorithmic fairness and transparency, safe and explainable AI, and privacy, among other topics,” says Miklau. “The group has since expanded to engage in related educational initiatives as well as to welcome faculty from across campus.”
Miklau’s own research focuses on balancing fairness and privacy. In a recent paper titled Fair Decision Making using Privacy-Protected Data, Miklau and collaborators from Duke and Colgate Universities investigated whether the “noise” added to data to protect individual privacy impacts the accuracy of the data in a biased way. U.S. Census data, for example, has a privacy mechanism applied to it, so that no individual can be identified through the data. Miklau and his collaborators showed that, for example, due to the “noise” added to U.S. county-level data in order to protect privacy, a county could be incorrectly identified as not having a sufficient Hispanic population, thus depriving voters in that county from receiving the dual-language election material mandated in the Voting Rights Act. Exploring topics like this can help governments and institutions reduce bias in their data-driven decisions.
Among other education activities, EQUATE is sponsoring COMPSCI 692E: Equitable Algorithms and Systems, a one-credit seminar that will give students the opportunity to learn more about EQUATE research across specialties. Students will take turns presenting an EQUATE-related paper from their respective research areas.
From computer vision to public policy to data diversity and more, CICS faculty are making strides in improving the outcome of algorithmic systems.
Visit the EQUATE website to learn more.