Regularized maximum correntropy machine

TL;DR

This paper introduces a regularized maximum correntropy framework for training classifiers robust to noisy labels, outperforming traditional methods by maximizing the similarity between predicted and true labels under regularization.

Contribution

It proposes a novel MCC-based learning algorithm that effectively handles noisy labels and controls model complexity, improving classification performance.

Findings

Significant performance improvement over traditional transitional loss methods.

Effective handling of noisy and outlying labels in classification tasks.

Demonstrated success on two challenging pattern classification datasets.

Abstract

In this paper we investigate the usage of regularized correntropy framework for learning of classifiers from noisy labels. The class label predictors learned by minimizing transitional loss functions are sensitive to the noisy and outlying labels of training samples, because the transitional loss functions are equally applied to all the samples. To solve this problem, we propose to learn the class label predictors by maximizing the correntropy between the predicted labels and the true labels of the training samples, under the regularized Maximum Correntropy Criteria (MCC) framework. Moreover, we regularize the predictor parameter to control the complexity of the predictor. The learning problem is formulated by an objective function considering the parameter regularization and MCC simultaneously. By optimizing the objective function alternately, we develop a novel predictor learning algorithm. The experiments on two chal- lenging pattern classification tasks show that it significantly outperforms the machines with transitional loss functions.