FAI: Fairness-Aware Algorithms for Network Analysis

点击翻译按钮获取中文摘要

As society becomes increasingly reliant on artificial intelligence (AI) technology, there have been growing concerns whether the decisions generated by the AI systems may lead to discriminatory actions against certain protected groups in the population. These concerns have brought increasing scrutiny into the issue of fairness in AI systems and their underlying machine learning algorithms. To overcome this challenge, the overarching goal of this project is to develop fairness-aware algorithms that can maintain the high utility of decisions generated by the AI systems without discriminating against particular subgroups of the population. Specifically, this research will address fundamental issues of fairness in algorithms that utilize network data in their decision making. Successful completion of this project will not only produce novel algorithms for researchers, but also tools that can help practitioners assess the level of inequality present in social networking platforms. The undergraduate and graduate students who participate in the project will be trained to conduct cutting edge research in AI and network science. The investigators will also seek collaborative partnership with research scientists from the industry to apply the developed methods in order to expand their social impact beyond the academic community.This research fills a major gap in current research on fairness in AI, which has primarily focused on independent and identically distributed (i.i.d.) data. There are still questions remain whether the existing methods are effective when applied to network data. In particular, the link structure of the network often contains information about the protected attributes (e.g., gender, race, or sexual orientation), and thus, must be taken into consideration in the design of fairness-aware machine learning algorithms. To address this issue, the objectives of this project are two-fold: (1) To develop metrics for assessing fairness in network learning algorithms and (2) To design, implement, and evaluate network learning algorithms that consider the tradeoff between fairness and utility of the models for various network analysis tasks and applications (including community detection and link prediction). The innovative methods developed in this project will be a step forward towards bridging the gap between current understanding of fairness in i.i.d. data and its application to network analysis.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.