Machine Learning (ML) systems, whether predictive or generative, not only reproduce biases and stereotypes but, even more worryingly, amplify them. Strategies for bias detection and mitigation typically focus on either ex post or ex ante approaches, but are always limited to two steps analyses. In this paper, we introduce the notion of Bias Amplification Chain (BAC) as a series of steps in which bias may be amplified during the design, development and deployment phases of trained models. We provide an application to such notion in the credit scoring setting and a quantitative analysis through the BRIO tool.
Bias Amplification Chains in ML-based Systems with an Application to Credit Scoring / A.G. Buda, G. Coraglia, F.A. Genco, C. Manganini, G. Primiero (CEUR WORKSHOP PROCEEDINGS). - In: BEWARE 2024 : Bias, Risk, Explainability, Ethical AI and the role of Logic and Logic Programming 2024 / [a cura di] G. Coraglia, F. A. D'Asaro, A. Dyoub, F. A. Lisi, G. Primiero. - [s.l] : CEUR-WS, 2024 Dec 22. - pp. 1-9 (( Intervento presentato al 3. convegno BEWARE 2024 : Bias, Risk, Explainability, Ethical AI and the role of Logic and Logic Programming 2024 tenutosi a Bolzano nel 2024.
Bias Amplification Chains in ML-based Systems with an Application to Credit Scoring
G. CoragliaSecondo
;F.A. Genco;C. ManganiniPenultimo
;G. PrimieroUltimo
2024
Abstract
Machine Learning (ML) systems, whether predictive or generative, not only reproduce biases and stereotypes but, even more worryingly, amplify them. Strategies for bias detection and mitigation typically focus on either ex post or ex ante approaches, but are always limited to two steps analyses. In this paper, we introduce the notion of Bias Amplification Chain (BAC) as a series of steps in which bias may be amplified during the design, development and deployment phases of trained models. We provide an application to such notion in the credit scoring setting and a quantitative analysis through the BRIO tool.File | Dimensione | Formato | |
---|---|---|---|
paper9.pdf
accesso aperto
Tipologia:
Publisher's version/PDF
Dimensione
1.13 MB
Formato
Adobe PDF
|
1.13 MB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.