FixOut: an ensemble approach to fairer models - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Pré-Publication, Document De Travail Année : 2020

FixOut: an ensemble approach to fairer models

Guilherme Alves
  • Fonction : Auteur
  • PersonId : 1083721
Vaishnavi Bhargava
  • Fonction : Auteur
  • PersonId : 1083722
Fabien Bernier
  • Fonction : Auteur
  • PersonId : 1083723
Miguel Couceiro
Amedeo Napoli

Résumé

In this paper, we address the question of process and model fairness. We propose FixOut, a human-centered and model-agnostic framework, that uses any explanation method (based on feature importance) to assess model's reliance on sensitive features. Given a pre-trained classifier, FixOut first checks whether it relies on user defined sensitive features. If it does, then FixOut employs feature dropout to produce a pool of simplified classifiers that are then aggregated into an ensemble classifier. We present empirical results using different models on several real-world datasets, that show a consistent improvement in terms of widely used fairness metrics, decreased reliance on sensitive features, and without compromising the classifier's accuracy.
Fichier principal
Vignette du fichier
_shared__IDA_2021___FixOut__an_ensemble_approach_to_fairer_models (1).pdf (362.4 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03033181 , version 1 (01-12-2020)

Identifiants

  • HAL Id : hal-03033181 , version 1

Citer

Guilherme Alves, Vaishnavi Bhargava, Fabien Bernier, Miguel Couceiro, Amedeo Napoli. FixOut: an ensemble approach to fairer models. 2020. ⟨hal-03033181⟩
232 Consultations
224 Téléchargements

Partager

Gmail Facebook X LinkedIn More