Searcharxiv⌕ Search

arXiv subjects

Alessia Berarducci

Publications and source records attributed to Alessia Berarducci.

2 recordsLinked to original sources

Causal Regularization for Robust Prediction

Prediction is an important use of a statistical model. However, predictive systems typically struggle when the environment, in which they were trained, changes. Robust prediction that can deal with distributional shifts has become an active research area. Existing approaches tend to make strong assumptions in order to achieve prediction guarantees. In this paper, we only require finite second moments in the training data to obtain sharp prediction guarantees. We generalize causal regularization to a framework that uses two environments and measures the differences between them through a generalized risk difference induced by changes in second-order moments of the covariates and response. By defining a sieve of out-of-sample distributions, we derive a worst-risk decomposition over increasing classes of environments and obtain a closed-form estimator that interpolates between OLS and the regularization limit solution. Causal interpretation is recovered under the inner-product invariance condition and when the conditional mean of the target given its causal parents remains invariant across environments. Our results show that the proposed estimator can achieve prediction stability over a broad class of perturbed environments, even when trained on a limited set. The framework separates two goals: robustness as a stable prediction under weak assumptions and causality as a stronger property obtained with additional invariance structure.

stat.ME↗

Automatic Causal Fairness Analysis with LLM-Generated Reporting

AutoML, intended as the process of automating the application of machine learning to real-world problems, is a key step for AI popularisation. Most AutoML frameworks are not accounting for the potential lack of fairness in the training data and in the corresponding predictions. We introduce \textsc{FairMind}, a software prototype aiming to automatise fairness analysis at the dataset level. We achieve that by resorting to the assumptions of the \emph{standard fairness model}, recently proposed by Plečko and Bareinboim. This allows for a sound fairness evaluation in terms of causal effects, based on \emph{counterfactual} queries involving the target, possibly confounders and mediators, and the different values of an input feature we regard as \emph{protected}. After the necessary data preprocessing, the tool implements a closed-form computation of the effects. LLMs are consequently exploited to generate accurate reports on the fairness levels detected in the training dataset. We achieve that in a zero-shot setup and show by examples the expected advantages with respect to a direct analysis performed by the LLM. To favour applications, extensions to ordinal protected variable and continuous targets and novel decomposition results are also discussed.

cs.LG↗