Showing papers by "Jean-Michel Loubes published in 2023"

On the coalitional decomposition of parameters of interest

Abstract: Understanding the behavior of a black-box model with probabilistic inputs can be based on the decomposition of a parameter of interest (e.g., its variance) into contributions attributed to each coalition of inputs (i.e., subsets of inputs). In this paper, we produce conditions for obtaining unambiguous and interpretable decompositions of very general parameters of interest. This allows to recover known decompositions, holding under weaker assumptions than stated in the literature.

When Mitigating Bias is Unfair: A Comprehensive Study on the Impact of Bias Mitigation Algorithms

Abstract: Most works on the fairness of machine learning systems focus on the blind optimization of common fairness metrics, such as Demographic Parity and Equalized Odds. In this paper, we conduct a comparative study of several bias mitigation approaches to investigate their behaviors at a fine grain, the prediction level. Our objective is to characterize the differences between fair models obtained with different approaches. With comparable performances in fairness and accuracy, are the different bias mitigation approaches impacting a similar number of individuals? Do they mitigate bias in a similar way? Do they affect the same individuals when debiasing a model? Our findings show that bias mitigation approaches differ a lot in their strategies, both in the number of impacted individuals and the populations targeted. More surprisingly, we show these results even apply for several runs of the same mitigation approach. These findings raise questions about the limitations of the current group fairness metrics, as well as the arbitrariness, hence unfairness, of the whole debiasing process.

How optimal transport can tackle gender biases in multi-class neural-network classifiers for job recommendations?

Abstract: Automatic recommendation systems based on deep neural networks have become extremely popular during the last decade. Some of these systems can, however, be used in applications that are ranked as High Risk by the European Commission in the AI act—for instance, online job candidate recommendations. When used in the European Union, commercial AI systems in such applications will be required to have proper statistical properties with regard to the potential discrimination they could engender. This motivated our contribution. We present a novel optimal transport strategy to mitigate undesirable algorithmic biases in multi-class neural network classification. Our strategy is model agnostic and can be used on any multi-class classification neural network model. To anticipate the certification of recommendation systems using textual data, we used it on the Bios dataset, for which the learning task consists of predicting the occupation of female and male individuals, based on their LinkedIn biography. The results showed that our approach can reduce undesired algorithmic biases in this context to lower levels than a standard strategy.

Counterfactual Explanation for Multivariate Times Series Using A Contrastive Variational Autoencoder

Abstract: We tackle the issue of anomaly detection for multivariate functional data in a supervised setting. Deep learning applied to multivariate time series has become common nowadays, especially for medical data such as electrocardiogram (ECG). There are not many explanability techniques that can handle multivariate time series. In this paper, we propose a model to understand abnormal class features on multivariate time series. We present a new variational autoencoder (VAE) training method that focuses on dividing the latent space into general and class-based features using supervised contrastive learning. The Contrastive VAE produces a well-organized latent space that enables us to modify only the class-based features and to use the generative part of the VAE to produce counterfactual examples. This method is able to easily provide plausible counterfactual observations, which highlights the differences between pathological and non-pathological data. We demonstrate the superiority of our approach over other counterfactual methods in terms of validity and performance.

Are fairness metric scores enough to assess discrimination biases in machine learning?

Abstract: This paper presents novel experiments shedding light on the shortcomings of current metrics for assessing biases of gender discrimination made by machine learning algorithms on textual data. We focus on the Bios dataset, and our learning task is to predict the occupation of individuals, based on their biography. Such prediction tasks are common in commercial Natural Language Processing (NLP) applications such as automatic job recommendations. We address an important limitation of theoretical discussions dealing with group-wise fairness metrics: they focus on large datasets, although the norm in many industrial NLP applications is to use small to reasonably large linguistic datasets for which the main practical constraint is to get a good prediction accuracy. We then question how reliable are different popular measures of bias when the size of the training set is simply sufficient to learn reasonably accurate predictions. Our experiments sample the Bios dataset and learn more than 200 models on different sample sizes. This allows us to statistically study our results and to confirm that common gender bias indices provide diverging and sometimes unreliable results when applied to relatively small training and test samples. This highlights the crucial importance of variance calculations for providing sound results in this field.