Advanced Seminar

Spatial Fairness in Algorithmic Decision Making: Concepts, Detection, and Mitigation

Organizers:
Dimitris Kyriakopoulos (Athena Research Center, Greece)
Dimitris Sacharidis (Université Libre de Bruxelles, Belgium)
Giorgos Giannopoulos (National Observatory of Athens and Athena Research Center, Greece)

Abstract: AI tools and applications are increasingly becoming part of many aspects of our lives, including social, economic, and professional domains. Going beyond functionalities such as simple recommendations, AI applications are also widely being used for decision assistance/support and decision-making. Combined with the largely black-box nature of current state-of-the-art algorithms, this has raised substantial concerns regarding the fairness of such algorithmic decision-making/support systems. To this end, a plethora of fairness and bias definitions have been proposed in the literature, along with methods for auditing, detecting, and mitigating bias. These methods operate either at the pre-processing stage (repairing the training data), the in-processing stage (intervening during the training of the AI model), or the post-processing stage (correcting the model’s decisions). In this seminar, we focus on spatial fairness, i.e., the notion that the distribution of a model’s decisions should not be affected (positively or negatively) by the location of the instances on which the decisions are made. We first provide a brief introduction to general-purpose AI fairness, including prominent fairness notions, definitions, and detection and mitigation techniques. We then proceed to discuss spatial fairness, presenting a categorization of its different interpretations and the types of techniques applied. We subsequently focus on the most prominent state-of-the-art methods for spatial bias detection and mitigation. The seminar concludes with a discussion of open challenges and future directions for research in this area.