Domain Detection
Domain detection focuses on identifying data points that originate from a distribution different from the model's training data, a crucial task for ensuring reliable model performance and safety in various applications. Current research emphasizes developing robust methods for detecting out-of-domain (OOD) data, exploring techniques like contrastive learning, K-nearest neighbor approaches, and Bayesian methods, often leveraging pre-trained language models or vision transformers. Successful domain detection is vital for improving the reliability and safety of AI systems across diverse fields, from automated driving and medical image analysis to conversational AI and social media monitoring.
Papers
Watch the Neighbors: A Unified K-Nearest Neighbor Contrastive Learning Framework for OOD Intent Discovery
Yutao Mou, Keqing He, Pei Wang, Yanan Wu, Jingang Wang, Wei Wu, Weiran Xu
Disentangling Confidence Score Distribution for Out-of-Domain Intent Detection with Energy-Based Learning
Yanan Wu, Zhiyuan Zeng, Keqing He, Yutao Mou, Pei Wang, Yuanmeng Yan, Weiran Xu