Search

Gli algoritmi sono vulnerabili a sviluppare pregiudizi che potrebbero rendere le loro decisioni ingiuste nei confronti di particolari gruppi di individui. La fairness comporta una serie di aspetti che dipendono fortemente dal dominio di applicazione che considera diverse nozioni su cosa sia una decisione giusta nelle situazioni che influiscono sugli individui della popolazione. Le precise differenze, implicazioni e ortogonalità tra queste nozioni non sono state ancora completamente analizzate e il lavoro prova a mattere ordini in questo zoo di definizioni. Quando si tratta di far rispettare tali vincoli, la maggior parte dei modelli di mitigazione in-processing incorpora vincoli di fairness come componente fondamentale della loss function e richiede quindi aggiustamenti a livello di codice per adattarsi a contesti specifici e domini. Piuttosto che fare affidamento su un approccio procedurale, il nostro modello sfrutta la struttura dichiarativa conoscenza acquisita per codificare i requisiti di equità sotto forma di regole logiche che catturano affermazioni del linguaggio naturale bigue e precise. Proponiamo un'integrazione neuro-simbolica basato su Logic Tensor Networks che combina apprendimento di reti nueurali basato sui daticonoscenze logiche di alto livello, consentendo di eseguire task di classificazione riducendo discriminazione. Le prove sperimentali mostrano che le prestazioni sono buone quanto quelle dello stato dell’arte fornendo così un quadro flessibile per tenere conto della non discriminazione, spesso a un costo modesto in termini di precisione., Algorithms are vulnerable to biases that might render their decisions unfair toward particular groups of individuals. Fairness comes with a range of facets that strongly depend on the application domain that consider different notions of what is a fair decision in situations impacting individuals in the population. The precise differences, implications and orthogonality between these notions have not yet been fully analyzed and we try to make some order out of this zoo of definitions. When it comes about enforcing such constraints, most in-processing mitigation models embed fairness constraints as fundamental component of the loss function thus requiring code-level adjustments to adapt to specific contexts and domains. Rather than relying on a procedural approach, our model leverages declarative structured knowledge to encode fairness requirements in the form of logic rules capturing unam- biguous and precise natural language statements. We propose a neuro-symbolic integration approach based on Logic Tensor Networks that combines data-driven network-based learning with high-level logical knowledge, allowing to perform classification tasks while reducing discrimination. Experimental evidence shows that performance is as good as state-of-the-art thus providing a flexible framework to account for non-discrimination often at a modest cost in terms of accuracy.

In recent years, the problem of addressing fairness in machine learning (ML) and automatic decision making has attracted a lot of attention in the scientific communities dealing with artificial intelligence. A plethora of different definitions of fairness in ML have been proposed, that consider different notions of what is a "fair decision" in situations impacting individuals in the population. The precise differences, implications and "orthogonality" between these notions have not yet been fully analyzed in the literature. In this work, we try to make some order out of this zoo of definitions.

Artificial intelligence (AI)-based decision-making systems are employed nowadays in an ever growing number of online as well as offline services-some of great importance. Depending on sophisticated learning algorithms and available data, these systems are increasingly becoming automated and data-driven. However, these systems can impact individuals and communities with ethical or legal consequences. Numerous approaches have therefore been proposed to develop decision-making systems that are discrimination-conscious by-design. However, these methods assume the underlying data distribution is stationary without drift, which is counterfactual in many realworld applications. In addition, their focus has been largely on minimizing discrimination while maximizing prediction performance without necessary flexibility in customizing the tradeoff according to different applications. To this end, we propose a learning algorithm for fair classification that also adapts to evolving data streams and further allows for a flexible control on the degree of accuracy and fairness. The positive results on a set of discriminated and non-stationary data streams demonstrate the effectiveness and flexibility of this approach.