The ability to detect Out-of-Domain (OOD) inputs has been a critical requirement in many real-world NLP applications since the inclusion of unsupported OOD inputs may lead to catastrophic failure of systems. However, it remains an empirical question whether current algorithms can tackle such problem 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 (ID) 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 over-confidence Error of Neural Networks, leading to more reliable ID and OOD prediction outcomes.

本研究提出了一个名为ProtoInfoMax的新型架构，可同时处理领域内和领域外句子，通过互信息最大化（InfoMax）目标来提高低资源文本分类中OOD检测的性能。实验结果表明，我们提出的方法可以显著提高20％的性能，并减少神经网络的典型自信错误，从而得到更可靠的预测结果。

ProtoInfoMax：利用互信息最大化的典型网络进行领域外检测