MetaAligner: Conditional Weak-to-Strong Correction for Generalizable Multi-Objective Alignment of Language Models.
Kailai YangZhiwei LiuQianqian XieTianlin ZhangNirui SongJimin HuangZiyan KuangSophia AnaniadouPublished in: CoRR (2024)
Keyphrases
- language model
- multi objective
- language modeling
- evolutionary algorithm
- document retrieval
- probabilistic model
- retrieval model
- language modelling
- n gram
- ad hoc information retrieval
- context sensitive
- language models for information retrieval
- speech recognition
- test collection
- information retrieval
- objective function
- query expansion
- smoothing methods
- query terms
- genetic algorithm
- vector space model
- statistical language models
- translation model
- passage retrieval
- term dependencies
- term frequency
- conditional probabilities