REVS: Unlearning Sensitive Information in Language Models via Rank Editing in the Vocabulary Space.
Tomer AshuachMartin TutekYonatan BelinkovPublished in: CoRR (2024)
Keyphrases
- language model
- sensitive information
- language modeling
- document retrieval
- third party
- privacy preserving
- spoken term detection
- speech recognition
- n gram
- out of vocabulary
- probabilistic model
- language modelling
- retrieval model
- query expansion
- privacy preservation
- statistical language models
- test collection
- language models for information retrieval
- data sets
- information retrieval
- pseudo relevance feedback
- context sensitive
- private information
- smoothing methods
- original data
- protect sensitive
- vector space model
- sensitive data
- confidential information
- pattern recognition
- relevance model
- document ranking
- raw data
- natural language processing
- training data