arXiv subjects

Eustasio del Barrio

Publications and source records attributed to Eustasio del Barrio.

27 records · Page 2Linked to original sources

arXiv2019

Box-constrained monotone $L_\infty$-approximations to Lipschitz regularizations, with applications to robust testing

Tests of fit to exact models in statistical analysis often lead to rejections even when the model is a useful approximate description of the random generator of the data. Among possible relaxations of a fixed model, the one defined by contamination neighbourhoods, namely, $\mathcal{V}_α(P_0)=\{(1-α)P_0+αQ: Q \in \mathcal{P}\}$, where $\mathcal{P}$ is the set of all probabilities in the sample space, has received much attention, from its central role in Robust Statistics. For probabilities on the real line, consistent tests of fit to $\mathcal{V}_α(P_0)$ can be based on $d_K(P_0,R_α(P))$, the minimal Kolmogorov distance between $P_0$ and the set of trimmings of $P$, $R_α(P)=\big\{\tilde P\in\mathcal{P}:\tilde P\ll P,\,{\textstyle \frac{d\tilde P}{dP}\leq\frac{1}{1-α}}\, P\text{-a.s.}\big\}$. We show that this functional admits equivalent formulations in terms of, either best approximation in uniform norm by $L$-Lipschitz functions satisfying a box constraint, or as the best monotone approximation in uniform norm to the $L$-Lipschitz regularization, which is seen to be expressable in terms of the average of the Pasch-Hausdorff envelopes. This representation for the solution of the variational problem allows to obtain results showing stability of the functional $d_K(P_0,R_α(P))$, as well as directional differentiability, providing the basis for a Central Limit Theorem for that functional.

math.OC↗

arXiv2019

On approximate validation of models: A Kolmogorov-Smirnov based approach

Classical tests of fit typically reject a model for large enough real data samples. In contrast, often in statistical practice a model offers a good description of the data even though it is not the "true" random generator. We consider a more flexible approach based on contamination neighbourhoods around a model. Using trimming methods and the Kolmogorov metric we introduce a functional statistic measuring departures from a contaminated model and the associated estimator corresponding to its sample version. We show how this estimator allows testing of fit for the (slightly) contaminated model vs sensible deviations from it, with uniformly exponentially small type I and type II error probabilities. We also address the asymptotic behavior of the estimator showing that, under suitable regularity conditions, it asymptotically behaves as the supremum of a Gaussian process. As an application we explore methods of comparison between descriptive models based on the paradigm of model falseness. We also include some connections of our approach with the False-Discovery-Rate setting, showing competitive behavior when estimating the contamination level, although applicable in a wider framework.

math.ST↗

arXiv2018

Obtaining fairness using optimal transport theory

Statistical algorithms are usually helping in making decisions in many aspects of our lives. But, how do we know if these algorithms are biased and commit unfair discrimination of a particular group of people, typically a minority? \textit{Fairness} is generally studied in a probabilistic framework where it is assumed that there exists a protected variable, whose use as an input of the algorithm may imply discrimination. There are different definitions of Fairness in the literature. In this paper we focus on two of them which are called Disparate Impact (DI) and Balanced Error Rate (BER). Both are based on the outcome of the algorithm across the different groups determined by the protected variable. The relationship between these two notions is also studied. The goals of this paper are to detect when a binary classification rule lacks fairness and to try to fight against the potential discrimination attributable to it. This can be done by modifying either the classifiers or the data itself. Our work falls into the second category and modifies the input data using optimal transport theory.

math.ST↗

arXiv2018

A Central Limit Theorem for $L_p$ transportation cost with applications to Fairness Assessment in Machine Learning

We provide a Central Limit Theorem for the Monge-Kantorovich distance between two empirical distributions with size $n$ and $m$, $W_p(P_n,Q_m)$ for $p>1$ for observations on the real line, using a minimal amount of assumptions. We provide an estimate of the asymptotic variance which enables to build a two sample test to assess the similarity between two distributions. This test is then used to provide a new criterion to assess the notion of fairness of a classification algorithm.

math.ST↗

arXiv2018

Confidence Intervals for Testing Disparate Impact in Fair Learning

We provide the asymptotic distribution of the major indexes used in the statistical literature to quantify disparate treatment in machine learning. We aim at promoting the use of confidence intervals when testing the so-called group disparate impact. We illustrate on some examples the importance of using confidence intervals and not a single value.

stat.ML↗

arXiv2015

Berry-Esseen bounds for weighted averages of Poisson avoidance functionals

We consider functionals which are weighted averages of the avoidance function of a Poisson process. Using the approach to Stein's method based on Malliavin calculus for Poisson functionals we provide explicit bounds for the Wasserstein distance between these standardized functionals and the standard normal distribution. Our approach relies on closed-form expressions for the action of some Malliavin type operators on avoidance functionals of Poisson processes. As a result we provide Berry-Esseen bounds in the CLT for the volume of the union of balls of a fixed radius around random Poisson centers or for the quantization error around points of a Poisson process. We also give Berry-Esseen bounds for avoidance functionals of empirical measures.

math.PR↗

arXiv2014

A contamination model for approximate stochastic order: extended version

Stochastic ordering among distributions has been considered in a variety of scenarios. Economic studies often involve research about the ordering of investment strategies or social welfare. However, as noted in the literature, stochastic orderings are often a too strong assumption which is not supported by the data even in cases in which the researcher tends to believe that a certain variable is somehow smaller than other. Instead of considering this rigid model of stochastic order we propose to look at a more flexible version in which two distributions are said to satisfy an approximate stochastic order relation if they are slightly contaminated versions of distributions which do satisfy the stochastic ordering. The minimal level of contamination that makes this approximate model hold can be used as a measure of the deviation of the original distributions from the exact stochastic order model. Our approach is based on the use of trimmings of probability measures. We discuss the connection between them and the approximate stochastic order model and provide theoretical support for its use in data analysis. We also provide simulation results.

stat.ME↗

arXiv2013

The empirical cost of optimal incomplete transportation

We consider the problem of optimal incomplete transportation between the empirical measure on an i.i.d. uniform sample on the d-dimensional unit cube $[0,1]^d$ and the true measure. This is a family of problems lying in between classical optimal transportation and nearest neighbor problems. We show that the empirical cost of optimal incomplete transportation vanishes at rate $O_P(n^{-1/d})$, where n denotes the sample size. In dimension $d\geq3$ the rate is the same as in classical optimal transportation, but in low dimension it is (much) higher than the classical rate.

math.PR↗

arXiv2012

Similarity of samples and trimming

We say that two probabilities are similar at level $α$ if they are contaminated versions (up to an $α$ fraction) of the same common probability. We show how this model is related to minimal distances between sets of trimmed probabilities. Empirical versions turn out to present an overfitting effect in the sense that trimming beyond the similarity level results in trimmed samples that are closer than expected to each other. We show how this can be combined with a bootstrap approach to assess similarity from two data samples.

math.ST↗