ProtoInfoMax: Prototypical Networks with Mutual Information Maximization for Out-of-Domain Detection
arXiv:2108.12229 · doi:10.18653/v1/2021.findings-emnlp.138
Abstract
The ability to detect Out-of-Domain (OOD) inputs has been a critical requirement in many real-world NLP applications. For example, intent classification in dialogue systems. The reason is that the inclusion of unsupported OOD inputs may lead to catastrophic failure of systems. However, it remains an empirical question whether current methods can tackle such problems reliably in a realistic scenario where zero OOD training data is available. In this study, we propose ProtoInfoMax, a new architecture that extends Prototypical Networks to simultaneously process in-domain and OOD sentences via Mutual Information Maximization (InfoMax) objective. Experimental results show that our proposed method can substantially improve performance up to 20% for OOD detection in low resource settings of text classification. We also show that ProtoInfoMax is less prone to typical overconfidence errors of Neural Networks, leading to more reliable prediction results.
arXiv commentThis manuscript will be available in ACL Anthology section EMNLP2021-Findings papers
References in corpus (10)
- Prototypical Networks for Few-shot Learning
- On Calibration of Modern Neural Networks
- Learning deep representations by mutual information estimation and maximization
- Recent Advances in Open Set Recognition: A Survey
- One-Class Classification: Taxonomy of Study and Review of Techniques
- Deep Anomaly Detection with Outlier Exposure
- On Mutual Information Maximization for Representation Learning
- A Mutual Information Maximization Perspective of Language Representation Learning
- Few-shot Text Classification with Distributional Signatures
- Discriminative Nearest Neighbor Few-Shot Intent Detection by Transferring Natural Language Inference