This paper presents new results on the Gelbrich distance and the corresponding ambiguity sets, to analyze the correlation between two scalar random variables. A closed expression of the minimum disturbance in the Gelbrich metric necessary to achieve a specified correlation between two random variables is proposed. This expression allows us to analy...
Unwanted bias is a major concern in machine learning, raising in particular significant ethical issues when machine learning models are deployed within high-stakes decision systems. A common solution to mitigate it is to integrate and optimize a statistical fairness metric along with accuracy during the training phase. However, one of the main rema...
Suppose we are given two datasets: a labeled dataset and unlabeled dataset which also has additional auxiliary features not present in the first dataset. What is the most principled way to use these datasets together to construct a predictor? The answer should depend upon whether these datasets are generated by the same or different distributions o...
Pour répondre aux enjeux de biais non-désirés en apprentissage machine, de nombreux travaux ont proposé des techniques d'amélioration de l'équité se basant sur des métriques statistiques. Cependant, l'expérience montre que la généralisation sur de nouvelles données n'est pas toujours au rendez-vous. Pour répondre à ce problème, nous proposons une t...
We study last-mile delivery with the option of crowd shipping, where a company makes use of occasional drivers to complement its vehicle’s fleet in the activity of delivering products to its customers. We model it as a data-driven distributionally robust optimization approach to the capacitated vehicle routing problem. We assume the marginals of th...
Published in Journal of machine learning research : JMLR
We present a Distributionally Robust Optimization (DRO) approach to estimate a robustified regression plane in a linear regression setting, when the observed samples are potentially contaminated with adversarially corrupted outliers. Our approach mitigates the impact of outliers by hedging against a family of probability distributions on the observ...
