Debmalya Mandal researches the design of large-scale voting systems that aggregate individual preferences and focuses on verifying and ensuring the fairness of AI systems. Before coming to Columbia, he completed a Ph.D. in computer science at Harvard University under the supervision of David C. Parkes. His thesis focused on designing systems that can help people in different fields improve their ability to make informed data-driven decisions. During his doctoral studies, he worked for the Chief Economist Office at Microsoft Research. Before his Ph.D., he received an M.E. in computer science from the Indian Institute of Science, Bangalore, and a B.E. in computer science from the Indian Institute of Engineering Science and Technology, Shibpur. In his spare time, he likes to read post-colonial literature, play soccer, and play classical guitar.

AG 1, AG 2, AG 3, INET, AG 4, AG 5, SWS, RG1, MMCI   Info

MPI Audience

English

-- Not specified --

Saarbrücken

One of the fundamental problems in multi-agent systems is voting i.e. aggregation of individual preferences. In the standard model of voting, voters express ranked preferences over alternatives and the voting rule aggregates them into a collective decision. In this talk, I will describe an unorthodox view of voting by expanding the design space to include both the elicitation rule, whereby voters map their (cardinal) preferences to votes and the aggregation rule, which transforms the reported votes into collective decisions. Intuitively, there is a tradeoff between the communication requirements of the elicitation rule and the efficiency of the outcome of the aggregation rule. I will provide an overview of the results that chart the Pareto frontier of this tradeoff.

In the second part of the talk, I will focus on the design of fair classifiers that are robust to perturbations in the training distribution. We will construct fair classifiers that are distributionally robust in the sense that their fairness guarantees hold even when the test distribution is different from the training distribution. Finally, if time permits, we will see the role of voting (or broadly social choice theory) in the design of fair algorithms.
Bio: Debmalya Mandal is a DSI postdoctoral fellow at Columbia University. Before joining Columbia, he received his Ph.D. in Computer Science from Harvard University, where he was advised by Prof. David C. Parkes. He is broadly interested in how AI systems can be integrated into society for the purpose of decision-making. In particular, his current interests include voting, information elicitation, and algorithmic fairness.

Danielle Dalton

+49 681 9303 9100

Zoom

994 5702 8566

passcode not visible

talk to your secretary

multi-agent systems

hosted by Goran Radanovic