Protecting People from Algorithms (and Vice Versa)

Computing technologies today have made it much easier to gather personal data. Algorithms are constantly analyzing such personal information and making consequential decisions on people. The extensive use of algorithms imposes the risks of algorithms mistreating people such as privacy violation or unfair discrimination. There is also a risk of people mistreating algorithms. For example, in a strategic environment people may have incentives to misreport their data to the algorithms for their own benefits. In this talk, I will first present an overarching theme in my research—protecting people and algorithms from each other. In particular, my work seeks to (1) protect people from algorithms in developing algorithms with privacy and fairness guarantees and (2) aims to protect algorithms from people in providing algorithms that incentivize people’s truthful behavior. I will then present two technical results in my work on differential privacy, a rigorous algorithmic notion for data privacy. The first result focuses on a fundamental problem in differential privacy—private query release. I will present a scalable algorithm that can accurately and privately answer a large collection of counting queries for high-dimensional data. In the second result, I will focus on a general framework for solving a family of economic optimization problems in a privacy-preserving manner. I will also demonstrate how differential privacy can be used as a novel tool to incentivize truth-telling when the algorithms need to elicit input data from self-interested participants.

No results