Beyond Mahalanobis-Based Scores for Textual OOD Detection
This addresses the vulnerability of NLP systems to distribution shifts in real-life applications, providing a practical tool for practitioners, though it appears incremental as it builds on existing OOD detection concepts.
The paper tackles the problem of detecting out-of-distribution (OOD) samples in NLP systems to prevent dysfunctions from distribution shifts, introducing TRUSTED, an unsupervised and fast OOD detector for Transformer-based classifiers that achieves state-of-the-art performance, improving previous AUROC by over 3 points.
Deep learning methods have boosted the adoption of NLP systems in real-life applications. However, they turn out to be vulnerable to distribution shifts over time which may cause severe dysfunctions in production systems, urging practitioners to develop tools to detect out-of-distribution (OOD) samples through the lens of the neural network. In this paper, we introduce TRUSTED, a new OOD detector for classifiers based on Transformer architectures that meets operational requirements: it is unsupervised and fast to compute. The efficiency of TRUSTED relies on the fruitful idea that all hidden layers carry relevant information to detect OOD examples. Based on this, for a given input, TRUSTED consists in (i) aggregating this information and (ii) computing a similarity score by exploiting the training distribution, leveraging the powerful concept of data depth. Our extensive numerical experiments involve 51k model configurations, including various checkpoints, seeds, and datasets, and demonstrate that TRUSTED achieves state-of-the-art performances. In particular, it improves previous AUROC over 3 points.