The Machine Learning Laboratory at Virginia Tech’s College of Engineering has received a $66,000 Amazon Research Award to develop algorithms to mitigate unfairness in recommendation engines.
Bert Huang, assistant professor of computer science at the Machine Learning Lab (MLL) and computer science PhD student Sirui Yao are working on developing algorithms to measure unfairness in machine learning programs, and to reduce the instances of this happening.
From the archive: Can technology aid in recruiting a diverse education workforce?
Such unfairness includes an engine that may not suggest computer science classes to a female student due to input data which represents an existing gender imbalance in the subject.
Huang said: “Recommendation engines can be unfair in that they may recommend beneficial products to some users but not others, or they may make more useful recommendations to some users than others.
“And the factors that determine which users get better quality recommendations may be based on irrelevant, unethical, or illegal discrimination.”
We are inspired by work in intersectionality, which tries to understand how different aspects of social identity intersect and exhibit different patterns of discrimination.
– Bert Huang, Machine Learning Lab
Huang also explained that other research in the area of machine learning unfairness is almost exclusively focused on situations where users’ group identities are known. This can be problematic because even when recommendation engines have information about group memberships, groups are not well defined, and there are always users who defy definitions.
He said: “Our team addresses a different setting where we do not know, or we do not trust, information about who belongs to what groups.”
The MLL team are working on algorithms that search over possible groups and try to make sure people are treated fairly across different group separations, such as gender.
Huang added: “We are inspired by work in intersectionality, which tries to understand how different aspects of social identity intersect and exhibit different patterns of discrimination.
“We aim to imbue algorithms with this same sense of multifaceted identity when considering algorithm fairness.”
More information on the Machine Learning Lab can be found at http://learning.cs.vt.edu/