Le 6e séminaire du Cycle de séminaires interdisciplinaires 2021-2022 Pôle SJPEG, Pôle AM2I, Pôle CLCS s’intitule :
« Making ML Models fairer through explanations and reinforcement »
Il sera donné par Miguel Couceiro, Professeur en Informatique, IDMC, LORIA, le 9 juin 2022 de 14 h à 16 h au LORIA (A008).
Algorithmic decisions are now being used on a daily basis, and based on Machine Learning (ML) processes that may be complex and biased. This raises several concerns given the critical impact that biased decisions may have on individuals or on society as a whole. Not only unfair outcomes affect human rights, they also undermine public trust in ML and AI. In this talk we will discuss approaches to mitigating unintended biases in ML models in order to make them fairer.
In the first part of this talk, we will address fairness issues of classifiers based on decision outcomes, and we will show how the simple idea of feature dropout followed by an ensemble approach can improve model fairness without compromising its accuracy. To illustrate we will present a general workflow that relies on explainers to tackle process fairness, which essentially measures a model’s reliance on sensitive or discriminatory features. We will present different applications and empirical settings that show improvements not only with respect to process fairness but also other fairness metrics.
In the second part of this talk, we will consider the particular case of masked language models and present a deep reinforcement learning approach to mitigating unintended biases such as gender, racial and demographic biases. As we will see, this task is rather complex and is far from being settled, and we will illustrate it with apparent success and hidden drawbacks.
Some of the results that we present constitute ongoing research being carried out in collaboration with Guilherme Alves, Vaishnavi Bhargava, Fabien Bernier, Luis Galarraga, Amedeo Napoli, Cindy Pereira and Rameez Querish.
Les intelligences artificielles et la cyber-sécurité soulèvent des questions mobilisant les méthodes propres à chacune des disciplines, dans des perspectives qui leurs sont propres. Ces questions sont ainsi propices à l’émergence de projets interdisciplinaires, sans qu’aucun des chercheurs impliqués ne soit réduit au rôle de prestataire de service. Il convient donc, par l’échange autour d’un animateur, de partager ce qui, dans les intelligences artificielles et la cyber-sécurité, suscite de véritables sujets de recherche à la croisée de plusieurs disciplines. L’objet de ces séminaires internes à nos laboratoires est de favoriser la présentation, en termes intelligibles, par un chercheur d’une discipline des aspects sur lesquels il travaille déjà et ceux sur lesquels il souhaiterait travailler en collaboration avec d’autres chercheurs d’autres disciplines.
Voici la liste de tous les séminaires de printemps 2022 qui auront lieu au Loria et sur Teams :
