Robust Label Shift Quantification

TL;DR

This paper introduces robust estimators for label shift quantification that align with maximum likelihood methods, offering theoretical guarantees and robustness against outliers, validated through analysis and empirical observations.

Contribution

It presents a novel robust estimation approach for label shift quantification that coincides with maximum likelihood, with theoretical analysis and robustness guarantees.

Findings

Proposed estimators match the MLE for label shift.

Derived deviation bounds with optimal guarantees.

Demonstrated robustness against outliers and contamination.

Abstract

In this paper, we investigate the label shift quantification problem. We propose robust estimators of the label distribution which turn out to coincide with the Maximum Likelihood Estimator. We analyze the theoretical aspects and derive deviation bounds for the proposed method, providing optimal guarantees in the well-specified case, along with notable robustness properties against outliers and contamination. Our results provide theoretical validation for empirical observations on the robustness of Maximum Likelihood Label Shift.