Deduplicating Training Data Mitigates Privacy Risks in Language Models.
Nikhil KandpalEric WallaceColin RaffelPublished in: ICML (2022)
Keyphrases
- language model
- training data
- language modeling
- n gram
- probabilistic model
- document retrieval
- information retrieval
- retrieval model
- language modelling
- classification accuracy
- decision trees
- supervised learning
- test collection
- training set
- speech recognition
- statistical language models
- labeled data
- vector space model
- context sensitive
- language model for information retrieval
- query expansion
- learning algorithm
- passage retrieval
- document length
- question answering
- pseudo relevance feedback
- document ranking
- ad hoc information retrieval
- retrieval effectiveness
- translation model
- smoothing methods
- web search
- language models for information retrieval