Jean-Michel Loubes

Traditional audit methods that rely on written records often miss the nuances of clinical reasoning that influence patient care. Ambient artificial intelligence captures spoken clinical encounters, allowing the analysis of real clinician–patient dialogue at scale. In a study of 124 urology consultations, a transcript-centered audit identified inter-physician variation and expert disagreement that conventional review missed. We explore the epistemic gains of this approach, its nonverbal blind spots, behavioral effects, technical vulnerabilities, and the EU AI Act’s regulatory landscape.

Optimal Transport (OT) is a resource allocation problem with applications in biology, data science, economics and statistics, among others. In some of the applications, practitioners have access to samples which approximate the continuous measure. Hence the quantities of interest derived from OT -- plans, maps and costs -- are only available in their empirical versions. Statistical inference on OT aims at finding confidence intervals of the population plans, maps and costs. In recent years this topic gained an increasing interest in the statistical community. In this paper we provide a comprehensive review of the most influential results on this research field, underlying the some of the applications. Finally, we provide a list of open problems.

We tackle the problem of bias mitigation of algorithmic decisions in a setting where both the output of the algorithm and the sensitive variable are continuous. Most of prior work deals with discrete sensitive variables, meaning that the biases are measured for subgroups of persons defined by a label, leaving out important algorithmic bias cases, where the sensitive variable is continuous. Typical examples are unfair decisions made with respect to the age or the financial status. In our work, we then propose a bias mitigation strategy for continuous sensitive variables, based on the notion of endogeneity which comes from the field of econometrics. In addition to solve this new problem, our bias mitigation strategy is a weakly supervised learning method which requires that a small portion of the data can be measured in a fair manner. It is model agnostic, in the sense that it does not make any hypothesis on the prediction model. It also makes use of a reasonably large amount of input observations and their corresponding predictions. Only a small fraction of the true output predictions should be known. This therefore limits the need for expert interventions. Results obtained on synthetic data show the effectiveness of our approach for examples as close as possible to real-life applications in econometrics.

Robustness studies of black-box models is recognized as a necessary task for numerical models based on structural equations and predictive models learned from data. These studies must assess the model’s robustness to possible misspecification of regarding its inputs (e.g., covariate shift). The study of black-box models, through the prism of uncertainty quantification (UQ), is often based on sensitivity analysis involving a probabilistic structure imposed on the inputs, while ML models are solely constructed from observed data. Our work aim at unifying the UQ and ML interpretability approaches, by providing relevant and easy-to-use tools for both paradigms. To provide a generic and understandable framework for robustness studies, we define perturbations of input information relying on quantile constraints and projections with respect to the Wasserstein distance between probability measures, while preserving their dependence structure. We show that this perturbation problem can be analytically solved. Ensuring regularity constraints by means of isotonic polynomial approximations leads to smoother perturbations, which can be more suitable in practice. Numerical experiments on real case studies, from the UQ and ML fields, highlight the computational feasibility of such studies and provide local and global insights on the robustness of black-box models to input perturbations.

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 recovering known decompositions, holding under weaker assumptions than the literature states.

The problem of algorithmic bias in machine learning has recently gained a lot of attention due to its potentially strong impact on our societies. In much the same manner, algorithmic biases can alter industrial and safety-critical machine learning applications, where high-dimensional inputs are used. This issue has, however, been mostly left out of the spotlight in the machine learning literature. Contrary to societal applications, where a set of potentially sensitive variables, such as gender or race, can be defined by common sense or by regulations to draw attention to potential risks, the sensitive variables are often unsuspected in industrial and safety-critical applications. In addition, these unsuspected sensitive variables may be indirectly represented as a latent feature of the input data. For instance, the predictions of an image classifier may be altered by reconstruction artefacts in a small subset of the training images. This raises serious and well-founded concerns about the commercial deployment of AI-based solutions, especially in a context where new regulations address bias issues in AI. The purpose of our paper is, then, to first give a large overview of recent advances in robust machine learning. Then, we propose a new procedure to detect and to treat such unknown biases. As far as we know, no equivalent procedure has been proposed in the literature so far. The procedure is also generic enough to be used in a wide variety of industrial contexts. Its relevance is demonstrated on a set of satellite images used to train a classifier. In this illustration, our technique detects that a subset of the training images has reconstruction faults, leading to systematic prediction errors that would have been unsuspected using conventional cross-validation techniques.

We prove a central limit theorem for the entropic transportation cost between subgaussian probability measures, centered at the population cost. This is the first result which allows for asymptotically valid inference for entropic optimal transport between measures which are not necessarily discrete. In the compactly supported case, we complement these results with new, faster, convergence rates for the expected entropic transportation cost between empirical measures. Our proof is based on strengthening convergence results for dual solutions to the entropic optimal transport problem.

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.

The diffeomorphic registration framework enables one to define an optimal matching function between two probability measures with respect to a data-fidelity loss function. The nonconvexity of the optimization problem renders the choice of this loss function crucial to avoid poor local minima. Recent work showed experimentally the efficiency of entropy-regularized optimal transportation costs, as they are computationally fast and differentiable while having few minima. Following this approach, we provide in this paper a new framework based on Sinkhorn divergences, unbiased entropic optimal transportation costs, and prove the statistical consistency with rate of the empirical optimal deformations.