Learning to classify with possible sensor failures
This work addresses robust classification for applications with unreliable sensors, but it is incremental as it builds on existing maximum entropy discrimination and regularization techniques.
The paper tackles the problem of learning a robust binary classifier in the presence of sensor failures that cause corrupt measurements in training data, aiming to minimize generalization error on non-corrupted data while controlling false alarm rates for anomalies. The proposed GEM-MED method demonstrates improved classification accuracy and anomaly detection rate over previous robust methods in simulated and real multimodal datasets.
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.