Калиброванная оценка уверенности

AbstractRecent studies in deep learning have shown significant progress in named entity recognition (NER). However, most existing works assume clean data annotation, while real-world scenarios typically involve a large amount …...

AbstractLarge pre-trained language models achieve impressive results across many tasks. However, recent works point out that pre-trained language models may memorize a considerable fraction of their training data, leading to …...