Utilizing Network Features to Detect Erroneous Inputs

Neural networks are vulnerable to a wide range of erroneous inputs such as corrupted, out-of-distribution, misclassified, and adversarial examples. Previously, separate solutions have been proposed for each of these faulty data types, however, in this work we show that a collective set of inputs with variegated data quality issues can be jointly identified with a single model. Specifically, we train a linear SVM classifier to detect four types of erroneous data using the hidden and softmax feature vectors of pre-trained neural networks. Our results indicate that these faulty data types generally exhibit linearly separable activation properties from correctly processed examples. We are able to identify erroneous inputs with an AUROC of 0.973 on CIFAR10, 0.957 on Tiny ImageNet, and 0.941 on ImageNet. We experimentally validate our findings across a diverse range of datasets, domains, and pre-trained models.