Same Pre-training Loss, Better Downstream: Implicit Bias Matters for Language Models.
Hong LiuSang Michael XieZhiyuan LiTengyu MaPublished in: ICML (2023)
Keyphrases
- language model
- language modeling
- n gram
- probabilistic model
- document retrieval
- speech recognition
- query expansion
- language modelling
- retrieval model
- information retrieval
- relevance model
- test collection
- smoothing methods
- statistical language models
- query terms
- vector space model
- context sensitive
- pseudo relevance feedback
- language models for information retrieval
- retrieval effectiveness
- word error rate
- ad hoc information retrieval
- query specific
- term dependencies
- translation model
- document length
- text categorization
- error rate