Regularized Best-of-N Sampling to Mitigate Reward Hacking for Language Model Alignment.
Yuu JinnaiTetsuro MorimuraKaito AriuKenshi AbePublished in: CoRR (2024)
Keyphrases
- language model
- language modeling
- probabilistic model
- n gram
- speech recognition
- retrieval model
- document retrieval
- ad hoc information retrieval
- information retrieval
- language modelling
- test collection
- query expansion
- reinforcement learning
- smoothing methods
- context sensitive
- vector space model
- pseudo relevance feedback
- word error rate
- statistical language models
- language model for information retrieval
- mixture model
- translation model
- machine learning