Sparse Autoencoders Find Highly Interpretable Features in Language Models.
Robert HubenHoagy CunninghamLogan RiggsAidan EwartLee SharkeyPublished in: ICLR (2024)
Keyphrases
- language model
- language modeling
- probabilistic model
- information retrieval
- n gram
- speech recognition
- document retrieval
- statistical language models
- query expansion
- retrieval model
- language modelling
- high dimensional
- language models for information retrieval
- denoising
- image features
- test collection
- semantic information
- vector space model
- document ranking
- term dependencies
- smoothing methods
- feature vectors