Adversarial Regression with Doubly Non-negative Weighting Matrices

Abstract

Many machine learning tasks that involve predicting an output response can be solved by training a weighted regression model. Unfortunately, the predictive power of this type of models may severely deteriorate under low sample sizes or under covariate perturbations. Reweighting the training samples has aroused as an effective mitigation strategy to these problems. In this paper, we propose a novel and coherent scheme for kernel-reweighted regression by reparametrizing the sample weights using a doubly non-negative matrix. When the weighting matrix is confined in an uncertainty set using either the log-determinant divergence or the Bures-Wasserstein distance, we show that the adversarially reweighted estimate can be solved efficiently using first-order methods. Numerical experiments show that our reweighting strategy delivers promising results on numerous datasets.

arXiv commentAccepted to the Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS2021)

References in corpus (6)

• Deep Speech 2: End-to-End Speech Recognition in English and Mandarin
• Equality of Opportunity in Supervised Learning
• From Parity to Preference-based Notions of Fairness in Classification
• Does Distributionally Robust Supervised Learning Give Robust Classifiers?
• Robustifying Conditional Portfolio Decisions via Optimal Transport
• Distributionally robust stochastic programs with side information based on trimmings -- Extended version