Syntactic Ghost: An Imperceptible General-purpose Backdoor Attacks on Pre-trained Language Models.
Pengzhou ChengWei DuZongru WuFengwei ZhangLibo ChenGongshen LiuPublished in: CoRR (2024)
Keyphrases
- language model
- pre trained
- watermarking scheme
- watermarking algorithm
- language modeling
- watermarking technique
- training data
- n gram
- probabilistic model
- document retrieval
- training examples
- speech recognition
- retrieval model
- test collection
- query expansion
- information retrieval
- context sensitive
- language modelling
- document ranking
- translation model
- control signals
- statistical language models
- language models for information retrieval
- natural language
- smoothing methods
- multi modal
- dimensionality reduction
- data mining
- neural network