Algorithms for Trustworthy AI

The development and diffusion of AI technology in recent years has been remarkable. However, various problems have also become apparent: the reliability problem, in which AI finds good solutions but we do not know why they are good, and the fairness problem, in which AI makes decisions that are unfavorable to people with certain attributes. Solving these problems requires efficient processing of a huge and complex search space. Therefore, our research aims to realize AI that truly meets society’s needs by using state-of-the-art algorithmic techniques. Specifically, we are studying fast algorithms using compressed data structures such as zero-suppressed binary decision diagrams (ZDDs). As applications to real-world problems, we are working on problems such as network reliability evaluation, fair shelter allocation, political redistricting, and diverse solution enumeration.