Fine-Tuning Deteriorates General Textual Out-of-Distribution Detection by Distorting Task-Agnostic Features.
Sishuo ChenWenkai YangXiaohan BiXu SunPublished in: EACL (Findings) (2023)
Keyphrases
- fine tuning
- special case
- spatial distribution
- false positives
- automatic detection
- textual features
- feature set
- image features
- low level
- classification accuracy
- probabilistic model
- object detection
- co occurrence
- detection algorithm
- prior knowledge
- feature extraction
- fine tune
- viable alternative
- power law
- anomaly detection
- detection rate
- benchmark datasets
- face detection
- computer vision
- multimedia
- metadata