ParaFuzz: An Interpretability-Driven Technique for Detecting Poisoned Samples in NLP.
Lu YanZhuo ZhangGuanhong TaoKaiyuan ZhangXuan ChenGuangyu ShenXiangyu ZhangPublished in: CoRR (2023)
Keyphrases
- natural language processing
- natural language
- training samples
- machine translation
- prediction accuracy
- data driven
- neural network
- sample set
- text analysis
- real time
- information extraction
- data mining
- sample points
- anomaly detection
- training examples
- supervised learning
- question answering
- automatic detection
- rule base
- data samples