Robust training on approximated minimal-entropy set

TL;DR

This paper introduces GEM-MED, a novel robust classification framework that jointly learns to classify and detect anomalies in data with corrupt measurements, improving accuracy and detection rates.

Contribution

The paper presents GEM-MED, a new non-parametric regularized maximum entropy method that enhances robustness in binary classification with anomalous training data.

Findings

GEM-MED outperforms previous methods in classification accuracy.

GEM-MED achieves higher anomaly detection rates.

The approach is validated on simulated and real multimodal datasets.

Abstract

In this paper, we propose a general framework to learn a robust large-margin binary classifier when corrupt measurements, called anomalies, caused by sensor failure might be present in the training set. The goal is to minimize the generalization error of the classifier on non-corrupted measurements while controlling the false alarm rate associated with anomalous samples. By incorporating a non-parametric regularizer based on an empirical entropy estimator, we propose a Geometric-Entropy-Minimization regularized Maximum Entropy Discrimination (GEM-MED) method to learn to classify and detect anomalies in a joint manner. We demonstrate using simulated data and a real multimodal data set. Our GEM-MED method can yield improved performance over previous robust classification methods in terms of both classification accuracy and anomaly detection rate.