NoiER: An Approach for Training more Reliable Fine-TunedDownstream Task Models
This addresses reliability issues in NLP downstream tasks for practitioners, though it is incremental as it builds on existing fine-tuning methods.
The paper tackles the distribution collapse problem in fine-tuned pretrained language models, where models fail to distinguish out-of-distribution sentences while being overconfident, and proposes noise entropy regularization (NoiER), which improves OOD detection metrics by 55% on average.
The recent development in pretrained language models trained in a self-supervised fashion, such as BERT, is driving rapid progress in the field of NLP. However, their brilliant performance is based on leveraging syntactic artifacts of the training data rather than fully understanding the intrinsic meaning of language. The excessive exploitation of spurious artifacts causes a problematic issue: The distribution collapse problem, which is the phenomenon that the model fine-tuned on downstream tasks is unable to distinguish out-of-distribution (OOD) sentences while producing a high confidence score. In this paper, we argue that distribution collapse is a prevalent issue in pretrained language models and propose noise entropy regularisation (NoiER) as an efficient learning paradigm that solves the problem without auxiliary models and additional~data. The proposed approach improved traditional OOD detection evaluation metrics by 55% on average compared to the original fine-tuned models.