Natural Attribute-based Shift Detection
This addresses reliability issues in deployed neural networks for practitioners in vision, language, and healthcare, but is incremental as it builds on existing OOD detection methods.
The paper tackles the problem of unpredictable behavior in neural networks on samples from distributions different than training by defining a new task called natural attribute-based shift (NAS) detection, and introduces benchmark datasets across vision, language, and medical domains, showing that prior OOD detection methods perform inconsistently on these datasets.
Despite the impressive performance of deep networks in vision, language, and healthcare, unpredictable behaviors on samples from the distribution different than the training distribution cause severe problems in deployment. For better reliability of neural-network-based classifiers, we define a new task, natural attribute-based shift (NAS) detection, to detect the samples shifted from the training distribution by some natural attribute such as age of subjects or brightness of images. Using the natural attributes present in existing datasets, we introduce benchmark datasets in vision, language, and medical for NAS detection. Further, we conduct an extensive evaluation of prior representative out-of-distribution (OOD) detection methods on NAS datasets and observe an inconsistency in their performance. To understand this, we provide an analysis on the relationship between the location of NAS samples in the feature space and the performance of distance- and confidence-based OOD detection methods. Based on the analysis, we split NAS samples into three categories and further suggest a simple modification to the training objective to obtain an improved OOD detection method that is capable of detecting samples from all NAS categories.