Sampling Ex-Post Fair Rankings
Abstract:
Rankings help users find the most relevant items quickly from a very large set of items. When the items to be ranked are people, news, media, products etc., unfairness in rankings can adversely impact opportunities, exposure, and worldview. Previous work on fair ranking has mostly focused on either ex-ante (in expectation) fairness in the output of randomized rankings or ex-post (actual) fairness in the output of deterministic rankings. I'll describe some recent results with my co-authors Sruthi Gorantla, Anay Mehrotra, and Anand Louis on randomized rankings that guarantee ex-post fairness. Any resemblance to algorithmic questions, new or old, for sampling random points from polytopes (Prof. Ravi Kannan's forte) is purely coincidental.
Speaker Bio
Amit Deshpande is a researcher at Microsoft Research India working
broadly in the areas of theoretical computer science and machine learning
with a focus on Algorithmic and mathematical techniques.
His current research interests are fairness and robustness of models
in supervised and unsupervised learning. He also works on sampling
techniques for subsampling large data to efficiently explore, summarize,
and learn. Amit completed his PhD from MIT in 2007 and prior to that
was did his Bachelors from Chennai Mathematical Institute, 2002.