Reinforcement Learning Fine-tuning of Language Models is Biased Towards More Extractable Features.
Diogo CruzEdoardo PonaAlex Holness-ToftsElias SchmiedVíctor Abia AlonsoCharlie GriffinBogdan-Ionut CirsteaPublished in: CoRR (2023)
Keyphrases
- language model
- fine tuning
- reinforcement learning
- language modeling
- n gram
- probabilistic model
- information retrieval
- language modelling
- document retrieval
- retrieval model
- statistical language models
- image features
- query expansion
- speech recognition
- test collection
- language models for information retrieval
- feature space
- document ranking
- feature vectors
- context sensitive
- vector space model
- co occurrence
- bayesian networks